Undergraduate Texts in Mathematics Editors
S. Axler K.A. Ribet
Undergraduate Texts in Mathematics Abbott: Understanding Analysis. Anglin: Mathematics: A Concise History and Philosophy. Readings in Mathematics. Anglin/Lambek: The Heritage of Thales. Readings in Mathematics. Apostol: Introduction to Analytic Number Theory. Second edition. Armstrong: Basic Topology. Armstrong: Groups and Symmetry. Axler: Linear Algebra Done Right. Second edition. Beardon: Limits: A New Approach to Real Analysis. Bak/Newman: Complex Analysis. Second edition. Banchoff/Wermer: Linear Algebra Through Geometry. Second edition. Berberian: A First Course in Real Analysis. Bix: Conics and Cubics: A Concrete Introduction to Algebraic Curves. Bre´maud: An Introduction to Probabilistic Modeling. Bressoud: Factorization and Primality Testing. Bressoud: Second Year Calculus. Readings in Mathematics. Brickman: Mathematical Introduction to Linear Programming and Game Theory. Browder: Mathematical Analysis: An Introduction. Buchmann: Introduction to Cryptography. Buskes/van Rooij: Topological Spaces: From Distance to Neighborhood. Callahan: The Geometry of Spacetime: An Introduction to Special and General Relavitity. Carter/van Brunt: The Lebesgue– Stieltjes Integral: A Practical Introduction. Cederberg: A Course in Modern Geometries. Second edition.
Chambert-Loir: A Field Guide to Algebra Childs: A Concrete Introduction to Higher Algebra. Second edition. Chung/AitSahlia: Elementary Probability Theory: With Stochastic Processes and an Introduction to Mathematical Finance. Fourth edition. Cox/Little/O’Shea: Ideals, Varieties, and Algorithms. Second edition. Croom: Basic Concepts of Algebraic Topology. Cull/Flahive/Robson: Difference Equations: From Rabbits to Chaos Curtis: Linear Algebra: An Introductory Approach. Fourth edition. Daepp/Gorkin: Reading, Writing, and Proving: A Closer Look at Mathematics. Devlin: The Joy of Sets: Fundamentals of Contemporary Set Theory. Second edition. Dixmier: General Topology. Driver: Why Math? Ebbinghaus/Flum/Thomas: Mathematical Logic. Second edition. Edgar: Measure, Topology, and Fractal Geometry. Elaydi: An Introduction to Difference Equations. Third edition. Erdo˜s/Sura´nyi: Topics in the Theory of Numbers. Estep: Practical Analysis in One Variable. Exner: An Accompaniment to Higher Mathematics. Exner: Inside Calculus. Fine/Rosenberger: The Fundamental Theory of Algebra. Fischer: Intermediate Real Analysis. Flanigan/Kazdan: Calculus Two: Linear and Nonlinear Functions. Second edition. Fleming: Functions of Several Variables. Second edition. Foulds: Combinatorial Optimization for Undergraduates. Foulds: Optimization Techniques: An Introduction. (continued on page 228)
John Stillwell
The Four Pillars of Geometry With 138 Illustrations
John Stillwell Department of Mathematics University of San Francisco San Francisco, CA 94117-1080 USA
[email protected]
Editorial Board S. Axler Mathematics Department San Francisco State University San Francisco, CA 94132 USA
K.A. Ribet Department of Mathematics University of California at Berkeley Berkeley, CA 94720-3840 USA
Mathematics Subject Classification (2000): 51-xx, 15-xx Library of Congress Control Number: 2005929630 ISBN-10: 0-387-25530-3 ISBN-13: 978-0387-25530-9
Printed on acid-free paper.
© 2005 Springer Science+Business Media, Inc. All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, Inc., 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Printed in the United States of America. 9 8 7 6 5 4 3 2 1 springeronline.com
(EB)
To Elaine
Preface Many people think there is only one “right” way to teach geometry. For two millennia, the “right” way was Euclid’s way, and it is still good in many respects. But in the 1950s the cry “Down with triangles!” was heard in France and new geometry books appeared, packed with linear algebra but with no diagrams. Was this the new “right” way, or was the “right” way something else again, perhaps transformation groups? In this book, I wish to show that geometry can be developed in four fundamentally different ways, and that all should be used if the subject is to be shown in all its splendor. Euclid-style construction and axiomatics seem the best way to start, but linear algebra smooths the later stages by replacing some tortuous arguments by simple calculations. And how can one avoid projective geometry? It not only explains why objects look the way they do; it also explains why geometry is entangled with algebra. Finally, one needs to know that there is not one geometry, but many, and transformation groups are the best way to distinguish between them. Two chapters are devoted to each approach: The first is concrete and introductory, whereas the second is more abstract. Thus, the first chapter on Euclid is about straightedge and compass constructions; the second is about axioms and theorems. The first chapter on linear algebra is about coordinates; the second is about vector spaces and the inner product. The first chapter on projective geometry is about perspective drawing; the second is about axioms for projective planes. The first chapter on transformation groups gives examples of transformations; the second constructs the hyperbolic plane from the transformations of the real projective line. I believe that students are shortchanged if they miss any of these four approaches to the subject. Geometry, of all subjects, should be about taking different viewpoints, and geometry is unique among the mathematical disciplines in its ability to look different from different angles. Some prefer vii
viii
Preface
to approach it visually, others algebraically, but the miracle is that they are all looking at the same thing. (It is as if one discovered that number theory need not use addition and multiplication, but could be based on, say, the exponential function.) The many faces of geometry are not only a source of amazement and delight. They are also a great help to the learner and teacher. We all know that some students prefer to visualize, whereas others prefer to reason or to calculate. Geometry has something for everybody, and all students will find themselves building on their strengths at some times, and working to overcome weaknesses at other times. We also know that Euclid has some beautiful proofs, whereas other theorems are more beautifully proved by algebra. In the multifaceted approach, every theorem can be given an elegant proof, and theorems with radically different proofs can be viewed from different sides. This book is based on the course Foundations of Geometry that I taught at the University of San Francisco in the spring of 2004. It should be possible to cover it all in a one-semester course, but if time is short, some sections or chapters can be omitted according to the taste of the instructor. For example, one could omit Chapter 6 or Chapter 8. (But with regret, I am sure!)
Acknowledgements My thanks go to the students in the course, for feedback on my raw lecture notes, and especially to Gina Campagna and Aaron Keel, who contributed several improvements. Thanks also go to my wife Elaine, who proofread the first version of the book, and to Robin Hartshorne, John Howe, Marc Ryser, Abe Shenitzer, and Michael Stillwell, who carefully read the revised version and saved me from many mathematical and stylistic errors. Finally, I am grateful to the M. C. Escher Company – Baarn – Holland for permission to reproduce the Escher work Circle Limit I shown in Figure 8.19, and the explicit mathematical transformation of it shown in Figure 8.10. This work is copyright (2005) The M. C. Escher Company. J OHN S TILLWELL San Francisco, November 2004 South Melbourne, April 2005
Contents Preface
vii
1 Straightedge and compass 1.1 Euclid’s construction axioms . . . . . . . . . . 1.2 Euclid’s construction of the equilateral triangle 1.3 Some basic constructions . . . . . . . . . . . . 1.4 Multiplication and division . . . . . . . . . . . 1.5 Similar triangles . . . . . . . . . . . . . . . . . 1.6 Discussion . . . . . . . . . . . . . . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
1 2 4 6 10 13 17
2 Euclid’s approach to geometry 2.1 The parallel axiom . . . . . . . . . 2.2 Congruence axioms . . . . . . . . . 2.3 Area and equality . . . . . . . . . . 2.4 Area of parallelograms and triangles 2.5 The Pythagorean theorem . . . . . . 2.6 Proof of the Thales theorem . . . . 2.7 Angles in a circle . . . . . . . . . . 2.8 The Pythagorean theorem revisited . 2.9 Discussion . . . . . . . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
20 21 24 26 29 32 34 36 38 42
. . . . . .
46 47 48 51 53 55 57
. . . . . . . . .
3 Coordinates 3.1 The number line and the number plane 3.2 Lines and their equations . . . . . . . 3.3 Distance . . . . . . . . . . . . . . . . 3.4 Intersections of lines and circles . . . 3.5 Angle and slope . . . . . . . . . . . . 3.6 Isometries . . . . . . . . . . . . . . . ix
. . . . . . . . .
. . . . . .
. . . . . . . . .
. . . . . .
. . . . . . . . .
. . . . . .
. . . . . . . . .
. . . . . .
. . . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
x
Contents 3.7 3.8
The three reflections theorem . . . . . . . . . . . . . . . . Discussion . . . . . . . . . . . . . . . . . . . . . . . . . .
61 63
4 Vectors and Euclidean spaces 4.1 Vectors . . . . . . . . . . . . . . . . . . 4.2 Direction and linear independence . . . . 4.3 Midpoints and centroids . . . . . . . . . 4.4 The inner product . . . . . . . . . . . . . 4.5 Inner product and cosine . . . . . . . . . 4.6 The triangle inequality . . . . . . . . . . 4.7 Rotations, matrices, and complex numbers 4.8 Discussion . . . . . . . . . . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
65 66 69 71 74 77 80 83 86
5 Perspective 5.1 Perspective drawing . . . . . . . . . . . . 5.2 Drawing with straightedge alone . . . . . 5.3 Projective plane axioms and their models 5.4 Homogeneous coordinates . . . . . . . . 5.5 Projection . . . . . . . . . . . . . . . . . 5.6 Linear fractional functions . . . . . . . . 5.7 The cross-ratio . . . . . . . . . . . . . . 5.8 What is special about the cross-ratio? . . 5.9 Discussion . . . . . . . . . . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
88 89 92 94 98 100 104 108 110 113
6 Projective planes 6.1 Pappus and Desargues revisited . . . 6.2 Coincidences . . . . . . . . . . . . 6.3 Variations on the Desargues theorem 6.4 Projective arithmetic . . . . . . . . 6.5 The field axioms . . . . . . . . . . 6.6 The associative laws . . . . . . . . 6.7 The distributive law . . . . . . . . . 6.8 Discussion . . . . . . . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
117 118 121 125 128 133 136 138 140
. . . .
143 144 146 151 154
. . . . . . . .
7 Transformations 7.1 The group of isometries of the plane . 7.2 Vector transformations . . . . . . . . 7.3 Transformations of the projective line 7.4 Spherical geometry . . . . . . . . . .
. . . . . . . .
. . . .
. . . . . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
Contents 7.5 7.6 7.7 7.8 7.9
xi The rotation group of the sphere . . . . . . Representing space rotations by quaternions A finite group of space rotations . . . . . . The groups S3 and RP3 . . . . . . . . . . . Discussion . . . . . . . . . . . . . . . . . .
8 Non-Euclidean geometry 8.1 Extending the projective line to a plane . 8.2 Complex conjugation . . . . . . . . . . 8.3 Reflections and M¨obius transformations 8.4 Preserving non-Euclidean lines . . . . . 8.5 Preserving angle . . . . . . . . . . . . . 8.6 Non-Euclidean distance . . . . . . . . . 8.7 Non-Euclidean translations and rotations 8.8 Three reflections or two involutions . . 8.9 Discussion . . . . . . . . . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
157 159 163 167 170
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
174 175 178 182 184 186 191 196 199 203
References
213
Index
215
1 Straightedge and compass P REVIEW For over 2000 years, mathematics was almost synonymous with the geometry of Euclid’s Elements, a book written around 300 BCE and used in school mathematics instruction until the 20th century. Euclidean geometry, as it is now called, was thought to be the foundation of all exact science. Euclidean geometry plays a different role today, because it is no longer expected to support everything else. “Non-Euclidean geometries” were discovered in the early 19th century, and they were found to be more useful than Euclid’s in certain situations. Nevertheless, non-Euclidean geometries arose as deviations from the Euclidean, so one first needs to know what they deviate from. A naive way to describe Euclidean geometry is to say it concerns the geometric figures that can be drawn (or constructed as we say) by straightedge and compass. Euclid assumes that it is possible to draw a straight line between any two given points, and to draw a circle with given center and radius. All of the propositions he proves are about figures built from straight lines and circles. Thus, to understand Euclidean geometry, one needs some idea of the scope of straightedge and compass constructions. This chapter reviews some basic constructions, to give a quick impression of the extent of Euclidean geometry, and to suggest why right angles and parallel lines play a special role in it. Constructions also help to expose the role of length, area, and angle in geometry. The deeper meaning of these concepts, and the related role of numbers in geometry, is a thread we will pursue throughout the book.
1
2
1 Straightedge and compass
1.1 Euclid’s construction axioms Euclid assumes that certain constructions can be done and he states these assumptions in a list called his axioms (traditionally called postulates). He assumes that it is possible to: 1. Draw a straight line segment between any two points. 2. Extend a straight line segment indefinitely. 3. Draw a circle with given center and radius. Axioms 1 and 2 say we have a straightedge, an instrument for drawing arbitrarily long line segments. Euclid and his contemporaries tried to avoid infinity, so they worked with line segments rather than with whole lines. This is no real restriction, but it involves the annoyance of having to extend line segments (or “produce” them, as they say in old geometry books). Today we replace Axioms 1 and 2 by the single axiom that a line can be drawn through any two points. The straightedge (unlike a ruler) has no scale marked on it and hence can be used only for drawing lines—not for measurement. Euclid separates the function of measurement from the function of drawing straight lines by giving measurement functionality only to the compass—the instrument assumed in Axiom 3. The compass is used to draw the circle through a given point B, with a given point A as center (Figure 1.1).
A B Figure 1.1: Drawing a circle
1.1 Euclid’s construction axioms
3
To do this job, the compass must rotate rigidly about A after being initially set on the two points A and B. Thus, it “stores” the length of the radius AB and allows this length to be transferred elsewhere. Figure 1.2 is a classic view of the compass as an instrument of measurement. It is William Blake’s painting of Isaac Newton as the measurer of the universe.
Figure 1.2: Blake’s painting of Newton the measurer The compass also enables us to add and subtract the length |AB| of AB from the length |CD| of another line segment CD by picking up the compass with radius set to |AB| and describing a circle with center D (Figure 1.3, also Elements, Propositions 2 and 3 of Book I). By adding a fixed length repeatedly, one can construct a “scale” on a given line, effectively creating a ruler. This process illustrates how the power of measuring lengths resides in the compass. Exactly which lengths can be measured in this way is a deep question, which belongs to algebra and analysis. The full story is beyond the scope of this book, but we say more about it below. Separating the concepts of “straightness” and “length,” as the straightedge and the compass do, turns out to be important for understanding the foundations of geometry. The same separation of concepts reappears in different approaches to geometry developed in Chapters 3 and 5.
4
1 Straightedge and compass
|A
B|
C
|CD| − |AB|
D |CD| + |AB|
Figure 1.3: Adding and subtracting lengths
1.2 Euclid’s construction of the equilateral triangle Constructing an equilateral triangle on a given side AB is the first proposition of the Elements, and it takes three steps: 1. Draw the circle with center A and radius AB. 2. Draw the circle with center B and radius AB. 3. Draw the line segments from A and B to the intersection C of the two circles just constructed. The result is the triangle ABC with sides AB, BC, and CA in Figure 1.4. C
A
B
Figure 1.4: Constructing an equilateral triangle Sides AB and CA have equal length because they are both radii of the first circle. Sides AB and BC have equal length because they are both radii of the second circle. Hence, all three sides of triangle ABC are equal.
1.2 Euclid’s construction of the equilateral triangle
5
This example nicely shows the interplay among • construction axioms, which guarantee the existence of the construction lines and circles (initially the two circles on radius AB and later the line segments BC and CA), • geometric axioms, which guarantee the existence of points required for later steps in the construction (the intersection C of the two circles), • and logic, which guarantees that certain conclusions follow. In this case, we are using a principle of logic that says that things equal to the same thing (both |BC| and |CA| equal |AB|) are equal to each other (so |BC| = |CA|). We have not yet discussed Euclid’s geometric axioms or logic. We use the same logic for all branches of mathematics, so it can be assumed “known,” but geometric axioms are less clear. Euclid drew attention to one and used others unconsciously (or, at any rate, without stating them). History has shown that Euclid correctly identified the most significant geometric axiom, namely the parallel axiom. We will see some reasons for its significance in the next section. The ultimate reason is that there are important geometries in which the parallel axiom is false. The other axioms are not significant in this sense, but they should also be identified for completeness, and we will do so in Chapter 2. In particular, it should be mentioned that Euclid states no axiom about the intersection of circles, so he has not justified the existence of the point C used in his very first proposition!
A question arising from Euclid’s construction The equilateral triangle is an example of a regular polygon: a geometric figure bounded by equal line segments that meet at equal angles. Another example is the regular hexagon in Exercise 1.2.1. If the polygon has n sides, we call it an n-gon, so the regular 3-gon and the regular 6-gon are constructible. For which n is the regular n-gon constructible? We will not completely answer this question, although we will show that the regular 4-gon and 5-gon are constructible. The question for general n turns out to belong to algebra and number theory, and a complete answer depends on a problem about prime numbers that has not yet been solved: m For which m is 22 + 1 a prime number?
6
1 Straightedge and compass
Exercises By extending Euclid’s construction of the equilateral triangle, construct: 1.2.1 A regular hexagon. 1.2.2 A tiling of the plane by equilateral triangles (solid lines in Figure 1.5). 1.2.3 A tiling of the plane by regular hexagons (dashed lines in Figure 1.5).
Figure 1.5: Triangle and hexagon tilings of the plane
1.3 Some basic constructions The equilateral triangle construction comes first in the Elements because several other constructions follow from it. Among them are constructions for bisecting a line segment and bisecting an angle. (“Bisect” is from the Latin for “cut in two.”) Bisecting a line segment To bisect a given line segment AB, draw the two circles with radius AB as above, but now consider both of their intersection points, C and D. The line CD connecting these points bisects the line segment AB (Figure 1.6).
1.3 Some basic constructions
7 C
B
A
D Figure 1.6: Bisecting a line segment AB Notice also that BC is perpendicular to AB, so this construction can be adapted to construct perpendiculars. • To construct the perpendicular to a line L at a point E on the line, first draw a circle with center E, cutting L at A and B. Then the line CD constructed in Figure 1.6 is the perpendicular through E. • To construct the perpendicular to a line L through a point E not on L , do the same; only make sure that the circle with center E is large enough to cut the line L at two different points. Bisecting an angle To bisect an angle POQ (Figure 1.7), first draw a circle with center O cutting OP at A and OQ at B. Then the perpendicular CD that bisects the line segment AB also bisects the angle POQ. P A C
D O
B
Q
Figure 1.7: Bisecting an angle POQ
8
1 Straightedge and compass
It seems from these two constructions that bisecting a line segment and bisecting an angle are virtually the same problem. Euclid bisects the angle before the line segment, but he uses two similar constructions (Elements, Propositions 9 and 10 of Book I). However, a distinction between line segments and angles emerges when we attempt division into three or more parts. There is a simple tool for dividing a line segment in any number of equal parts—parallel lines—but no corresponding tool for dividing angles. Constructing the parallel to a line through a given point We use the two constructions of perpendiculars noted above—for a point off the line and a point on the line. Given a line L and a point P outside L , first construct the perpendicular line M to L through P. Then construct the perpendicular to M through P, which is the parallel to L through P. Dividing a line segment into n equal parts Given a line segment AB, draw any other line L through A and mark n successive, equally spaced points A1 , A2 , A3 , . . . , An along L using the compass set to any fixed radius. Figure 1.8 shows the case n = 5. Then connect An to B, and draw the parallels to BAn through A1 , A2 , . . . , An−1 . These parallels divide AB into n equal parts. L A5 A4 A3 A2 A1 B
A Figure 1.8: Dividing a line segment into equal parts
This construction depends on a property of parallel lines sometimes attributed to Thales (Greek mathematician from around 600 BCE): parallels cut any lines they cross in proportional segments. The most commonly used instance of this theorem is shown in Figure 1.9, where a parallel to one side of a triangle cuts the other two sides proportionally.
1.3 Some basic constructions
9
The line L parallel to the side BC cuts side AB into the segments AP and PB, side AC into AQ and QC, and |AP|/|PB| = |AQ|/|QC|. A L
P
Q
B
C
Figure 1.9: The Thales theorem in a triangle This theorem of Thales is the key to using algebra in geometry. In the next section we see how it may be used to multiply and divide line segments, and in Chapter 2 we investigate how it may be derived from fundamental geometric principles.
Exercises 1.3.1 Check for yourself the constructions of perpendiculars and parallels described in words above. 1.3.2 Can you find a more direct construction of parallels? Perpendiculars give another important polygon—the square. 1.3.3 Give a construction of the square on a given line segment. 1.3.4 Give a construction of the square tiling of the plane. One might try to use division of a line segment into n equal parts to divide an angle into n equal parts as shown in Figure 1.10. We mark A on OP and B at equal distance on OQ as before, and then try to divide angle POQ by dividing line segment AB. However, this method is faulty even for division into three parts.
P
B
A
Q
O Figure 1.10: Faulty trisection of an angle 1.3.5 Explain why division of AB into three equal parts (trisection) does not always divide angle POQ into three equal parts. (Hint: Consider the case in which POQ is nearly a straight line.)
10
1 Straightedge and compass
The version of the Thales theorem given above (referring to Figure 1.9) has an equivalent form that is often useful. 1.3.6 If A, B,C, P, Q are as in Figure 1.9, so that |AP|/|PB| = |AQ|/|QC|, show that this equation is equivalent to |AP|/|AB| = |AQ|/|AC|.
1.4 Multiplication and division Not only can one add and subtract line segments (Section 1.1); one can also multiply and divide them. The product ab and quotient a/b of line segments a and b are obtained by the straightedge and compass constructions below. The key ingredients are parallels, and the key geometric property involved is the Thales theorem on the proportionality of line segments cut off by parallel lines. To get started, it is necessary to choose a line segment as the unit of length, 1, which has the property that 1a = a for any length a. Product of line segments To multiply line segment b by line segment a, we first construct any triangle UOA with |OU| = 1 and |OA| = a. We then extend OU by length b to B 1 and construct the parallel to UA through B1 . Suppose this parallel meets the extension of OA at C (Figure 1.11). By the Thales theorem, |AC| = ab. B1
b
Multiply by a
U 1 O
ab
a A
Figure 1.11: The product of line segments
C
1.4 Multiplication and division
11
Quotient of line segments To divide line segment b by line segment a, we begin with the same triangle UOA with |OU| = 1 and |OA| = a. Then we extend OA by distance b to B2 and construct the parallel to UA through B2 . Suppose that this parallel meets the extension of OU at D (Figure 1.12). By the Thales theorem, |UD| = b/a. D
b/a
Divide by a
U 1 O
b
a A
B2
Figure 1.12: The quotient of line segments The sum operation from Section 1.1 allows us to construct a segment n units in length, for any natural number n, simply by adding the segment 1 to itself n times. The quotient operation then allows us to construct a segment of length m/n, for any natural numbers m and n = 0. These are what we call the rational lengths. A great discovery of the Pythagoreans was that some lengths are not rational, and that some of these “irrational” lengths can be constructed by straightedge and compass. It is not known how the Pythagoreans made this discovery, but it has a connection with the Thales theorem, as we will see in the next section.
Exercises Exercise 1.3.6 showed that if PQ is parallel to BC in Figure 1.9, then |AP|/|AB| = |AQ|/|AC|. That is, a parallel implies proportional (left and right) sides. The following exercise shows the converse: proportional sides imply a parallel, or (equivalently), a nonparallel implies nonproportional sides. 1.4.1 Using Figure 1.13, or otherwise, show that if PR is not parallel to BC, then |AP|/|AB| = |AR|/|AC|.
12
1 Straightedge and compass A L
Q
P
R B
C
Figure 1.13: Converse of the Thales theorem 1.4.2 Conclude from Exercise 1.4.1 that if P is any point on AB and Q is any point on AC, then PQ is parallel to BC if and only if |AP|/|AB| = |AQ|/|AC|. The “only if” direction of Exercise 1.4.2 leads to two famous theorems—the Pappus and Desargues theorems—that play an important role in the foundations of geometry. We will meet them in more general form later. In their simplest form, they are the following theorems about parallels. 1.4.3 (Pappus of Alexandria, around 300 CE) Suppose that A, B,C, D, E, F lie alternately on lines L and M as shown in Figure 1.14.
M C
E A
O
L F
B
D
Figure 1.14: The parallel Pappus configuration Use the Thales theorem to show that if AB is parallel to ED and FE is parallel to BC then |OA| |OC| = . |OF| |OD| Deduce from Exercise 1.4.2 that AF is parallel to CD.
1.5 Similar triangles
13
1.4.4 (Girard Desargues, 1648) Suppose that points A, B,C, A , B ,C lie on concurrent lines L , M , N as shown in Figure 1.15. (The triangles ABC and A BC are said to be “in perspective from O.”)
L A A B
O
B
M
C C
N
Figure 1.15: The parallel Desargues configuration Use the Thales theorem to show that if AB is parallel to A B and BC is parallel to BC , then |OA| |OA | = . |OC| |OC | Deduce from Exercise 1.4.2 that AC is parallel to AC .
1.5 Similar triangles Triangles ABC and A BC are called similar if their corresponding angles are equal, that is, if angle at A = angle at A (= α say), angle at B = angle at B (= β say), angle at C = angle at C (= γ say). It turns out that equal angles imply that all sides are proportional, so we may say that one triangle is a magnification of the other, or that they have the same “shape.” This important result extends the Thales theorem, and actually follows from it.
14
1 Straightedge and compass
Why similar triangles have proportional sides Imagine moving triangle ABC so that vertex A coincides with A and sides AB and AC lie on sides A B and AC , respectively. Then we obtain the situation shown in Figure 1.16. In this figure, b and c denote the side lengths of triangle ABC opposite vertices B and C, respectively, and b and c denote the side lengths of triangle A BC (= ABC ) opposite vertices B and C , respectively.
c−
B β
c
B β c
A = A
γ
α
γ C
b
b − b
C
Figure 1.16: Similar triangles Because BC and BC both meet AB at angle β , they are parallel, and so it follows from the Thales theorem (Section 1.3) that b b − b = . c c −c Multiplying both sides by c(c − c) gives b(c − c) = c(b − b), that is, bc − bc = cb − cb, and hence
bc = cb .
Finally, dividing both sides by cc , we get b b = . c c That is, corresponding sides of triangles ABC and A BC opposite to the angles β and γ are proportional.
1.5 Similar triangles
15
We got this result by making the angles α in the two triangles coincide. If we make the angles β coincide instead, we similarly find that the sides opposite to α and γ are proportional. Thus, in fact, all corresponding sides of similar triangles are proportional. This consequence of the Thales theorem has many implications. In everyday life, it underlies the existence of scale maps, house plans, engineering drawings, and so on. In pure geometry, its implications are even more varied. Here is just one, which shows why square roots and irrational numbers turn up in geometry. The diagonal of the unit square is
√ 2
The diagonals of the unit square cut it into four quarters, each of which is a triangle similar to the half square cut off by a diagonal (Figure 1.17).
d/ 2
2 d/ d/ 2
2 d/
1 Figure 1.17: Quarters and halves of the square Each of the triangles in question has one right angle and two half right angles, so it follows from the theorem above that corresponding sides of any two of these triangles are proportional. In particular, if we take the half square, with short side 1 and long side d, and compare it with the quarter square, with short side d/2 and long side 1, we get short 1 d/2 = = . long d 1 Multiplying both sides of the equation by 2d gives 2 = d 2 , so d =
√ 2.
16
1 Straightedge and compass
√ The great, but disturbing, discovery of the Pythagoreans is√that 2 is irrational. That is, there are no natural numbers m and n such 2 = m/n. If there are such m and n we√can assume that they have no common divisor, and then the assumption 2 = m/n implies 2 = m2 /n2 2
2
squaring both sides multiplying both sides by n2
hence
m = 2n
hence
m2 is even
hence
m is even
since the square of an odd number is odd
hence
m = 2l
for some natural number l
hence
m2 = 4l 2 = 2n2
hence
n2 = 2l 2
hence
n2 is even
hence
n is even
since the square of an odd number is odd.
Thus, m and n have the common divisor 2, contrary to assumption. Our original assumption √ is therefore false, so there are no natural numbers m and n such that 2 = m/n. Lengths, products, and area Geometry obviously has to include the diagonal of the unit square, hence geometry includes the study of irrational lengths. This discovery troubled the ancient Greeks, because they did not believe that irrational lengths could be treated like numbers. In particular, the idea of interpreting the product of line segments as another line segment is not in Euclid. It first appears in Descartes’ G´eom´etrie of 1637, where algebra is used systematically in geometry for the first time. The Greeks viewed the product of line segments a and b as the rectangle with perpendicular sides a and b. If lengths are not necessarily numbers, then the product of two lengths is best interpreted as an area, and the product of three lengths as a volume—but then the product of four lengths seems to have no meaning at all. This difficulty perhaps explains why algebra appeared comparatively late in the development of geometry. On the other hand, interpreting the product of lengths as an area gives some remarkable insights, as we will see in Chapter 2. So it is also possible that algebra had to wait until the Greek concept of product had exhausted its usefulness.
1.6 Discussion
17
Exercises In general, two geometric figures are called similar if one is a magnification of the long side other. Thus, two rectangles are similar if the ratio short side is the same for both.
√ 2+1
1
√ 2−1
1
Figure 1.18: A pair of similar rectangles √ 2+1 1
1.5.1 Show that are similar.
=
√1 2−1
and hence that the two rectangles in Figure 1.18
1.5.2 Deduce that if a rectangle with long side a and short side b has the same shape as the two above, then so has the rectangle with long side b and short side a − 2b. √ This simple observation gives another proof that 2 is irrational: √ 1.5.3 Suppose that 2 + 1 = m/n, where m and n are natural numbers √ with m as small as possible. Deduce from Exercise 1.5.2 that we also have 2 + 1 = n/(m − 2n). This is a contradiction. Why? √ 1.5.4 It follows √ from Exercise 1.5.3 that 2+1 is irrational. Why does this imply that 2 is irrational?
1.6 Discussion Euclid’s Elements is the most influential book in the history of mathematics, and anyone interested in geometry should own a copy. It is not easy reading, but you will find yourself returning to it year after year and noticing something new. The standard edition in English is Heath’s translation, which is now available as a Dover reprint of the 1925 Cambridge University Press edition. This reprint is carried by many bookstores; I have even seen it for sale at Los Angeles airport! Its main drawback is its size—three bulky volumes—due to the fact that more than half the content consists of
18
1 Straightedge and compass
Heath’s commentary. You can find the Heath translation without the commentary in the Britannica Great Books of the Western World, Volume 11. These books can often be found in used bookstores. Another, more recent, one-volume edition of the Heath translation is Euclid’s Elements, edited by Dana Densmore and published by Green Lion Press in 2003. A second (slight) drawback of the Heath edition is that it is about 80 years old and beginning to sound a little antiquated. Heath’s English is sometimes quaint, and his commentary does not draw on modern research in geometry. He does not even mention some important advances that were known to experts in 1925. For this reason, a modern version of the Elements is desirable. A perfect version for the 21st century does not yet exist, but there is a nice concise web version by David Joyce at http://aleph0.clarkeu.edu/~djoyce/java/elements/elements.html
This Elements has a small amount of commentary, but I mainly recommend it for proofs in simple modern English and nice diagrams. The diagrams are “variable” by dragging points on the screen, so each diagram represents all possible situations covered by a theorem. For modern commentary on Euclid, I recommend two books: Euclid: the Creation of Mathematics by Benno Artmann and Geometry: Euclid and Beyond by Robin Hartshorne, published by Springer-Verlag in 1999 and 2000, respectively. Both books take Euclid as their starting point. Artmann mainly fills in the Greek background, although he also takes care to make it understandable to modern readers. Hartshorne is more concerned with what came after Euclid, and he gives a very thorough analysis of the gaps in Euclid and the ways they were filled by modern mathematicians. You will find Hartshorne useful supplementary reading for Chapters 2 and 3, where we examine the logical structure of the Elements and some of its gaps. The climax of the Elements is the theory of regular polyhedra in Book XIII. Only five regular polyhedra exist, and they are shown in Figure 1.19. Notice that three of them are built from equilateral triangles, one from squares, and one from regular pentagons. This remarkable phenomenon underlines the importance of equilateral triangles and squares, and draws attention to the regular pentagon. In Chapter 2, we show how to construct it. Some geometers believe that the material in the Elements was chosen very much with the theory of regular polyhedra in mind. For example, Euclid wants to construct the equilateral triangle, square, and pentagon in order to construct the regular polyhedra.
1.6 Discussion
19
Cube
Dodecahedron
Octahedron
Icosahedron
Tetrahedron
Figure 1.19: The regular polyhedra It is fortunate that Euclid did not need regular polygons more complex than the pentagon, because none were constructed until modern times. The regular 17-gon was constructed by the 19-year-old Carl Friedrich Gauss in 1796, and his discovery was the key to the “question arising” from the construction of the equilateral triangle in Section 1.2: for which n is the regular n-gon constructible? Gauss showed (with some steps filled in by Pierre Wantzel in 1837) that a regular polygon with a prime number p of m sides is constructible just in case p is of the form 22 + 1. This result gives three constructible p-gons not known to the Greeks, because 24 + 1 = 17,
28 + 1 = 257,
216 + 1 = 65537 m
are all prime numbers. But no larger prime numbers of the form 2 2 + 1 are known! Thus we do not know whether a larger constructible p-gon exists. These results show that the Elements is not all of geometry, even if one accepts the same subject matter as Euclid. To see where Euclid fits in the general panorama of geometry, I recommend the books Geometry and the Imagination by D. Hilbert and S. Cohn-Vossen, and Introduction to Geometry by H. S. M. Coxeter (Wiley, 1969).
2 Euclid’s approach to geometry P REVIEW Length is the fundamental concept of Euclid’s geometry, but several important theorems seem to be “really” about angle or area—for example, the theorem on the sum of angles in a triangle and the Pythagorean theorem on the sum of squares. Also, Euclid often uses area to prove theorems about length, such as the Thales theorem. In this chapter, we retrace some of Euclid’s steps in the theory of angle and area to show how they lead to the Pythagorean theorem and the Thales theorem. We begin with his theory of angle, which shows most clearly the influence of his parallel axiom, the defining axiom of what is now called Euclidean geometry. Angle is linked with length from the beginning by the so-called SAS (“side angle side”) criterion for equal triangles (or “congruent triangles,” as we now call them). We observe the implications of SAS for isosceles triangles and the properties of angles in a circle, and we note the related criterion, ASA (“angle side angle”). The theory of area depends on ASA, and it leads directly to a proof of the Pythagorean theorem. It leads more subtly to the Thales theorem and its consequences that we saw in Chapter 1. The theory of angle then combines nicely with the Thales theorem to give a second proof of the Pythagorean theorem. In following these deductive threads, we learn more about the scope of straightedge and compass constructions, partly in the exercises. Interesting spinoffs from these investigations include a process for cutting any polygon into pieces that form a square, a construction for the square root of any length, and a construction of the regular pentagon.
20
2.1 The parallel axiom
21
2.1 The parallel axiom In Chapter 1, we saw how useful it is to have rectangles: four-sided polygons whose angles are all right angles. Rectangles owe their existence to parallel lines—lines that do not meet—and fundamentally to the parallel axiom that Euclid stated as follows. Euclid’s parallel axiom. If a straight line crossing two straight lines makes the interior angles on one side together less than two right angles, then the two straight lines will meet on that side. Figure 2.1 illustrates the situation described by Euclid’s parallel axiom, which is what happens when the two lines are not parallel. If α + β is less than two right angles, then L and M meet somewhere on the right. N
β
M
L
α
Figure 2.1: When lines are not parallel It follows that if L and M do not meet on either side, then α + β = π . In other words, if L and M are parallel, then α and β together make a straight angle and the angles made by L , M , and N are as shown in Figure 2.2. N
α π −α
π −α α
Figure 2.2: When lines are parallel
M
L
22
2 Euclid’s approach to geometry
It also follows that any line through the intersection of N and M , not meeting L , makes the angle π − α with N . Hence, this line equals M . That is, if a parallel to L through a given point exists, it is unique. It is a little more subtle to show the existence of a parallel to L through a given point P, but one way is to appeal to a principle called ASA (“angle side angle”), which will be discussed in Section 2.2. Suppose that the lines L , M , and N make angles as shown in Figure 2.2, and that L and M are not parallel. Then, on at least one side of N , there is a triangle whose sides are the segment of N between L and M and the segments of L and M between N and the point where they meet. According to ASA, this triangle is completely determined by the angles α , π − α and the segment of N between them. But then an identical triangle is determined on the other side of N , and hence L and M also meet on the other side. This result contradicts Euclid’s assumption (implicit in the construction axioms discussed in Section 1.1) that there is a unique line through any two points. Hence, the lines L and M are in fact parallel when the angles are as shown in Figure 2.2. Thus, both the existence and the uniqueness of parallels follow from Euclid’s parallel axiom (existence “follows trivially,” because Euclid’s parallel axiom is not required). It turns out that they also imply it, so the parallel axiom can be stated equivalently as follows. Modern parallel axiom. For any line L and point P outside L , there is exactly one line through P that does not meet L . This form of the parallel axiom is often called “Playfair’s axiom,” after the Scottish mathematician John Playfair who used it in a textbook in 1795. Playfair’s axiom is simpler in form than Euclid’s, because it does not involve angles, and this is often convenient. However, we often need parallel lines and the equal angles they create, the so-called alternate interior angles (for example, the angles marked α in Figure 2.2). In such situations, we prefer to use Euclid’s parallel axiom.
Angles in a triangle The existence of parallels and the equality of alternate interior angles imply a beautiful property of triangles. Angle sum of a triangle. If α , β , and γ are the angles of any triangle, then α + β + γ = π .
2.1 The parallel axiom
23
To prove this property, draw a line L through one vertex of the triangle, parallel to the opposite side, as shown in Figure 2.3. L
α
α
β
γ
γ
Figure 2.3: The angle sum of a triangle Then the angle on the left beneath L is alternate to the angle α in the triangle, so it is equal to α . Similarly, the angle on the right beneath L is equal to γ . But then the straight angle π beneath L equals α + β + γ , the angle sum of the triangle.
Exercises The triangle is the most important polygon, because any polygon can be built from triangles. For example, the angle sum of any quadrilateral (polygon with four sides) can be worked out by cutting the quadrilateral into two triangles. 2.1.1 Show that the angle sum of any quadrilateral is 2π . A polygon P is called convex if the line segment between any two points in P lies entirely in P. For these polygons, it is also easy to find the angle sum. 2.1.2 Explain why a convex n-gon can be cut into n − 2 triangles. 2.1.3 Use the dissection of the n-gon into triangles to show that the angle sum of a convex n-gon is (n − 2)π . 2.1.4 Use Exercise 2.1.3 to find the angle at each vertex of a regular n-gon (an n-gon with equal sides and equal angles). 2.1.5 Deduce from Exercise 2.1.4 that copies of a regular n-gon can tile the plane only for n = 3, 4, 6.
24
2 Euclid’s approach to geometry
2.2 Congruence axioms Euclid says that two geometric figures coincide when one of them can be moved to fit exactly on the other. He uses the idea of moving one figure to coincide with another in the proof of Proposition 4 of Book I: If two triangles have two corresponding sides equal, and the angles between these sides equal, then their third sides and the corresponding two angles are also equal. His proof consists of moving one triangle so that the equal angles of the two triangles coincide, and the equal sides as well. But then the third sides necessarily coincide, because their endpoints do, and hence, so do the other two angles. Today we say that two triangles are congruent when their corresponding angles and side lengths are equal, and we no longer attempt to prove the proposition above. Instead, we take it as an axiom (that is, an unproved assumption), because it seems simpler to assume it than to introduce the concept of motion into geometry. The axiom is often called SAS (for “side angle side”). SAS axiom. If triangles ABC and A BC are such that |AB| = |A B |,
angle ABC = angle A BC ,
|BC| = |BC |
then also |AC| = |AC |,
angle BCA = angle BC A ,
angle CAB = angle C A B .
For brevity, one often expresses SAS by saying that two triangles are congruent if two sides and the included angle are equal. There are similar conditions, ASA and SSS, which also imply congruence (but SSA does not—can you see why?). They can be deduced from SAS, so it is not necessary to take them as axioms. However, we will assume ASA here to save time, because it seems just as natural as SAS. One of the most important consequences of SAS is Euclid’s Proposition 5 of Book I. It says that a triangle with two equal sides has two equal angles. Such a triangle is called isosceles, from the Greek for “equal sides.” The spectacular proof below is not from Euclid, but from the Greek mathematician Pappus, who lived around 300 CE. Isosceles triangle theorem. If a triangle has two equal sides, then the angles opposite to these sides are also equal.
2.2 Congruence axioms
25
Suppose that triangle ABC has |AB| = |AC|. Then triangles ABC and ACB, which of course are the same triangle, are congruent by SAS (Figure 2.4). Their left sides are equal, their right sides are equal, and so are the angles between their left and right sides, because they are the same angle (the angle at A). A
B
A
C
C
B
Figure 2.4: Two views of an isosceles triangle But then it follows from SAS that all corresponding angles of these triangles are equal: for example, the bottom left angles. In other words, the angle at B equals the angle at C, so the angles opposite to the equal sides are equal. A useful consequence of ASA is the following theorem about parallelograms, which enables us to determine the area of triangles. (Remember, a parallelogram is defined as a figure bounded by two pairs of parallel lines—the definition does not say anything about the lengths of its sides.) Parallelogram side theorem. Opposite sides of a parallelogram are equal. To prove this theorem we divide the parallelogram into triangles by a diagonal (Figure 2.5), and try to prove that these triangles are congruent. They are, because • they have the common side AC, • their corresponding angles α are equal, being alternate interior angles for the parallels AD and BC, • their corresponding angles β are equal, being alternate interior angles for the parallels AB and DC.
26
2 Euclid’s approach to geometry D
C
β
α
α
β
A
B
Figure 2.5: Dividing a parallelogram into triangles Therefore, the triangles are congruent by ASA, and in particular we have the equalities |AB| = |DC| and |AD| = |BC| between corresponding sides. But these are also the opposite sides of the parallelogram.
Exercises 2.2.1 Using the parallelogram side theorem and ASA, find congruent triangles in Figure 2.6. Hence, show that the diagonals of a parallelogram bisect each other.
Figure 2.6: A parallelogram and its diagonals 2.2.2 Deduce that the diagonals of a rhombus—a parallelogram whose sides are all equal—meet at right angles. (Hint: You may find it convenient to use SSS, which says that triangles are congruent when their corresponding sides are equal.) 2.2.3 Prove the isosceles triangle theorem differently by bisecting the angle at A.
2.3 Area and equality The principle of logic used in Section 1.2—that things equal to the same thing are equal to each other—is one of five principles that Euclid calls common notions. The common notions he states are particularly important for his theory of area, and they are as follows:
2.3 Area and equality
27
1. Things equal to the same thing are also equal to one another. 2. If equals are added to equals, the wholes are equal. 3. If equals are subtracted from equals, the remainders are equal. 4. Things that coincide with one another are equal to one another. 5. The whole is greater than the part. The word “equal” here means “equal in some specific respect.” In most cases, it means “equal in length” or “equal in area,” although Euclid’s idea of “equal in area” is not exactly the same as ours, as I will explain below. Likewise, “addition” can mean addition of lengths or addition of areas, but Euclid never adds a length to an area because this has no meaning in his system. A simple but important example that illustrates the use of “equals” is Euclid’s Proposition 15 of Book I: Vertically opposite angles are equal. Vertically opposite angles are the angles α shown in Figure 2.7.
α
β
α
Figure 2.7: Vertically opposite angles They are equal because each of them equals a straight angle minus β .
The square of a sum Proposition 4 of Book II is another interesting example. It states a property of squares and rectangles that we express by the algebraic formula (a + b)2 = a2 + 2ab + b2 . Euclid does not have algebraic notation, so he has to state this equation in words: If a line is cut at random, the square on the whole is equal to the squares on the segments and twice the rectangle contained by the segments. Whichever way you say it, Figure 2.8 explains why it is true. The line is a + b because it is cut into the two segments a and b, and hence
28
2 Euclid’s approach to geometry
b
ab
b2
a
a2
ab
a
b
Figure 2.8: The square of a sum of line segments • The square on the line is what we write as (a + b)2 . • The squares on the two segments a and b are a2 and b2 , respectively. • The rectangle “contained” by the segments a and b is ab. • The square (a+b)2 equals (in area) the sum of a2 , b2 , and two copies of ab. It should be emphasized that, in Greek mathematics, the only interpretation of ab, the “product” of line segments a and b, is the rectangle with perpendicular sides a and b (or “contained in” a and b, as Euclid used to say). This rectangle could be shown “equal” to certain other regions, but only by cutting the regions into identical pieces by straight lines. The Greeks did not realize that this “equality of regions” was the same as equality of numbers—the numbers we call the areas of the regions—partly because they did not regard irrational lengths as numbers, and partly because they did not think the product of lengths should be a length. As mentioned in Section 1.5, this belief was not necessarily an obstacle to the development of geometry. To find the area of nonrectangular regions, such as triangles or parallelograms, one has to think about cutting regions into pieces in any case. For such simple regions, there is no particular advantage in thinking of the area as a number, as we will see in Section 2.4. But first we need to investigate the concept mentioned in Euclid’s Common Notion number 4. What does it mean for one figure to “coincide” with another?
2.4 Area of parallelograms and triangles
29
Exercises In Figure 2.8, the large square is subdivided by two lines: one of them perpendicular to the bottom side of the square and the other perpendicular to the left side of the square. 2.3.1 Use the parallel axiom to explain why all other angles in the figure are necessarily right angles. Figure 2.8 presents the algebraic identity (a + b)2 = a2 + 2ab + b2 in geometric form. Other well-known algebraic identities can also be given a geometric presentation. 2.3.2 Give a diagram for the identity a(b + c) = ab + ac. 2.3.3 Give a diagram for the identity a2 − b2 = (a + b)(a − b). Euclid does not give a geometric theorem that explains the identity (a + b)3 = a3 + 3a2 b + 3ab2 + b3 . But it is not hard to do so by interpreting (a + b)3 as a cube with edge length a + b, a3 as a cube with edge a, a2 b as a box with perpendicular edges a, a, and b, and so on. 2.3.4 Draw a picture of a cube with edges a+b, and show it cut by planes (parallel to its faces) that divide each edge into a segment of length a and a segment of length b. 2.3.5 Explain why these planes cut the original cube into eight pieces: • a cube with edges a, • a cube with edges b, • three boxes with edges a, a, b, • three boxes with edges a, b, b.
2.4 Area of parallelograms and triangles The first nonrectangular region that can be shown “equal” to a rectangle in Euclid’s sense is a parallelogram. Figure 2.9 shows how to use straight lines to cut a parallelogram into pieces that can be reassembled to form a rectangle.
=
=
Figure 2.9: Assembling parallelogram and rectangle from the same pieces
30
2 Euclid’s approach to geometry
Only one cut is needed in the example of Figure 2.9, but more cuts are needed if the parallelogram is more sheared, as in Figure 2.10. 3
3
=
2
2
1
1
Figure 2.10: A case in which more cuts are required In Figure 2.10 we need two cuts, which produce the pieces labeled 1, 2, 3. The number of cuts can become arbitrarily large as the parallelogram is sheared further. We can avoid large numbers of cuts by allowing subtraction of pieces as well as addition. Figure 2.11 shows how to convert any rectangle to any parallelogram with the same base OR and the same height OP. We need only add a triangle, and then subtract an equal triangle. P
O
Q
S
T
R
Figure 2.11: Rectangle and parallelogram with the same base and height To be precise, if we start with rectangle OPQR and add triangle RQT , then subtract triangle OPS (which equals triangle RQT by the parallelogram side theorem of Section 2.2), the result is parallelogram OST R. Thus, the parallelogram is equal (in area) to a rectangle with the same base and height. We write this fact as area of parallelogram = base × height. To find the area of a triangle ABC, we notice that it can be viewed as “half” of a parallelogram by adding to it the congruent triangle ACD as shown in Figure 2.5, and again in Figure 2.12. D
A
C
B
Figure 2.12: A triangle as half a parallelogram
2.4 Area of parallelograms and triangles
31
Clearly, area of triangle ABC +area of triangle ACD = area of parallelogram ABCD, and the two triangles “coincide” (because they are congruent) and so they have equal area by Euclid’s Common Notion 4. Thus, area of triangle =
1 base × height. 2
This formula is important in two ways: • As a statement about area. From a modern viewpoint, the formula gives the area of the triangle as a product of numbers. From the ancient viewpoint, it gives a rectangle “equal” to the triangle, namely, the rectangle with the same base and half the height of the triangle. • As a statement about proportionality. For triangles with the same height, the formula shows that their areas are proportional to their bases. This statement turns out to be crucial for the proof of the Thales theorem (Section 2.6). The proportionality statement follows from the assumption that each line segment has a real number length, which depends on the acceptance of irrational numbers. As mentioned in the previous section, the Greeks did not accept this assumption. Euclid got the proportionality statement by a lengthy and subtle “theory of proportion” in Book V of the Elements.
Exercises To back up the claim that the formula 12 base × height gives a way to find the area of the triangle, we should explain how to find the height. 2.4.1 Given a triangle with a particular side specified as the “base,” show how to find the height by straightedge and compass construction. The equality of triangles OPS and RQT follows from the parallelogram side theorem, as claimed above, but a careful proof would explain what other axioms are involved. 2.4.2 By what Common Notion does |PQ| = |ST |? 2.4.3 By what Common Notion does |PS| = |QT |? 2.4.4 By what congruence axiom is triangle OPS congruent to triangle RQT ?
32
2 Euclid’s approach to geometry
2.5 The Pythagorean theorem The Pythagorean theorem is about areas, and indeed Euclid proves it immediately after he has developed the theory of area for parallelograms and triangles in Book I of the Elements. First let us recall the statement of the theorem. Pythagorean theorem. For any right-angled triangle, the sum of the squares on the two shorter sides equals the square on the hypotenuse. We follow Euclid’s proof, in which he divides the square on the hypotenuse into the two rectangles shown in Figure 2.13. He then shows that the light gray square equals the light gray rectangle and that the dark gray square equals the dark gray rectangle, so the sum of the light and dark squares is the square on the hypotenuse, as required.
Figure 2.13: Dividing the square for Euclid’s proof First we show equality for the light gray regions in Figure 2.13, and in fact we show that half of the light gray square equals half of the light gray rectangle. We start with a light gray triangle that is obviously half of the light gray square, and we successively replace it with triangles of the same base or height, ending with a triangle that is obviously half of the light gray rectangle (Figure 2.14).
2.5 The Pythagorean theorem
33
Start with half of the light gray square
Same base (side of light gray square) and height
Congruent triangle, by SAS (the included angle is the sum of the same parts)
Same base (side of square on hypotenuse) and height; new triangle is half the light gray rectangle
Figure 2.14: Changing the triangle without changing its area The same argument applies to the dark gray regions, and thus, the Pythagorean theorem is proved. Figure 2.13 suggests a natural way to construct a square equal in area to a given rectangle. Given the light gray rectangle, say, the problem is to reconstruct the rest of Figure 2.13.
34
2 Euclid’s approach to geometry
We can certainly extend a given rectangle to a square and hence reconstruct the square on the hypotenuse. The main problem is to reconstruct the right-angled triangle, from the hypotenuse, so that the other vertex lies on the dashed line. See whether you can think of a way to do this; a really elegant solution is given in Section 2.7. Once we have the right-angled triangle, we can certainly construct the squares on its other two sides—in particular, the gray square equal in area to the gray rectangle.
Exercises It follows from the Pythagorean theorem that a right-angled triangle with sides 3 √ √ and 4 has hypotenuse 32 + 42 = 25 = 5. But there is only one triangle with sides 3, 4, and 5 (by the SSS criterion mentioned in Exercise 2.2.2), so putting together lengths 3, 4, and 5 always makes a right-angled triangle. This triangle is known as the (3, 4, 5) triangle. 2.5.1 Verify that the (5, 12, 13), (8, 15, 17), and (7, 24, 25) triangles are rightangled. 2.5.2 Prove the converse Pythagorean theorem: If a, b, c > 0 and a2 + b2 = c2 , then the triangle with sides a, b, c is right-angled. 2.5.3 How can we be sure that lengths a, b, c > 0 with a2 + b2 = c2 actually fit together to make a triangle? (Hint: Show that a + b > c.) Right-angled triangles can be used to construct certain irrational lengths. For example, we√saw in Section 1.5 that the right-angled triangle with sides 1, 1 has hypotenuse 2. √ 2.5.4 Starting from the triangle√with sides 1, 1, and 2, find a straightedge and compass construction of 3. √ 2.5.5 Hence, obtain constructions of n for n = 2, 3, 4, 5, 6, . . ..
2.6 Proof of the Thales theorem We mentioned this theorem in Chapter 1 as a fact with many interesting consequences, such as the proportionality of similar triangles. We are now in a position to prove the theorem as Euclid did in his Proposition 2 of Book VI. Here again is a statement of the theorem. The Thales theorem. A line drawn parallel to one side of a triangle cuts the other two sides proportionally.
2.6 Proof of the Thales theorem
35
The proof begins by considering triangle ABC, with its sides AB and AC cut by the parallel PQ to side BC (Figure 2.15). Because PQ is parallel to BC, the triangles PQB and PQC on base PQ have the same height, namely the distance between the parallels. They therefore have the same area. A
Q
P
B
C
Figure 2.15: Triangle sides cut by a parallel If we add triangle APQ to each of the equal-area triangles PQB and PQC, we get the triangles AQB and APC, respectively. Hence, the latter triangles are also equal in area. Now consider the two triangles—APQ and PQB—that make up triangle AQB as triangles with bases on the line AB. They have the same height relative to this base (namely, the perpendicular distance of Q from AB). Hence, their bases are in the ratio of their areas: |AP| area APQ = . |PB| area PQB Similarly, considering the triangles APQ and PQC that make up the triangle APC, we find that |AQ| area APQ = . |QC| area PQC Because area PQB equals area PQC, the right sides of these two equations are equal, and so are their left sides. That is, |AP| |AQ| = . |PB| |QC| In other words, the line PQ cuts the sides AB and AC proportionally.
36
2 Euclid’s approach to geometry
Exercises As seen in Exercise 1.3.6, |AP|/|PB| = |AQ|/|QC| is equivalent to |AP|/|AB| = |AQ|/|AC|. This equation is a more convenient formulation of the Thales theorem if you want to prove the following generalization: 2.6.1 Suppose that there are several parallels P1 Q1 , P2 Q2 , P3 Q3 . . . to the side BC of triangle ABC. Show that |AP1 | |AP2 | |AP3 | |AB| = = = ··· = . |AQ1 | |AQ2 | |AQ3 | |AC| We can also drop the assumption that the parallels P1 Q1 , P2 Q2 , P3 Q3 . . . fall across a triangle ABC. 2.6.2 If parallels P1 Q1 , P2 Q2 , P3 Q3 . . . fall across a pair of parallel lines L and M , what can we say about the lengths they cut from L and M ?
2.7 Angles in a circle The isosceles triangle theorem of Section 2.2, simple though it is, has a remarkable consequence. Invariance of angles in a circle. If A and B are two points on a circle, then, for all points C on one of the arcs connecting them, the angle ACB is constant. To prove invariance we draw lines from A, B,C to the center of the circle, O, along with the lines making the angle ACB (Figure 2.16). Because all radii of the circle are equal, |OA| = |OC|. Thus triangle AOC is isosceles, and the angles α in it are equal by the isosceles triangle theorem. The angles β in triangle BOC are equal for the same reason. Because the angle sum of any triangle is π (Section 2.1), it follows that the angle at O in triangle AOC is π − 2α and the angle at O in triangle BOC is π − 2β . It follows that the third angle at O, angle AOB, is 2(α + β ), because the total angle around any point is 2π . But angle AOB is constant, so α + β is also constant, and α + β is precisely the angle at C. An important special case of this theorem is when A, O, and B lie in a straight line, so 2(α + β ) = π . In this case, α + β = π /2, and thus we have the following theorem (which is also attributed to Thales). Angle in a semicircle theorem. If A and B are the ends of a diameter of a circle, and C is any other point on the circle, then angle ACB is a right angle.
2.7 Angles in a circle
37
C β α
π − 2α π − 2β
O 2(α + β )
β
B α
A
Figure 2.16: Angle α + β in a circle
This theorem enables us to solve the problem left open at the end of Section 2.5: Given a hypotenuse AB, how do we construct the right-angled triangle whose other vertex C lies on a given line? Figure 2.17 shows how.
C
A
B
Figure 2.17: Constructing a right-angled triangle with given hypotenuse
38
2 Euclid’s approach to geometry
The trick is to draw the semicircle on diameter AB, which can be done by first bisecting AB to obtain the center of the circle. Then the point where the semicircle meets the given line (shown dashed) is necessarily the other vertex C, because the angle at C is a right angle. This construction completes the solution of the problem raised at the end of Section 2.5: finding a square equal in area to a given rectangle. In Section 2.8 we will show that Figure 2.17 also enables us to construct the square root of an arbitrary length, and it gives a new proof of the Pythagorean theorem.
Exercises 2.7.1 Explain how the angle in a semicircle theorem enables us to construct a right-angled triangle with a given hypotenuse AB. 2.7.2 Then, by looking at Figure 2.13 from the bottom up, find a way to construct a square equal in area to a given rectangle. 2.7.3 Given any two squares, we can construct a square that equals (in area) the sum of the two given squares. Why? 2.7.4 Deduce from the previous exercises that any polygon may be “squared”; that is, there is a straightedge and compass construction of a square equal in area to the given polygon. (You may assume that the given polygon can be cut into triangles.) The possibility of “squaring” any polygon was apparently known to Greek mathematicians, and this may be what tempted them to try “squaring the circle”: constructing a square equal in area to a given circle. There is no straightedge and compass solution of the latter problem, but this was not known until 1882. Coming back to angles in the circle, here is another theorem about invariance of angles: 2.7.5 If a quadrilateral has its vertices on a circle, show that its opposite angles sum to π .
2.8 The Pythagorean theorem revisited In Book VI, Proposition 31 of the Elements, Euclid proves a generalization of the Pythagorean theorem. From it, we get a new proof of the ordinary Pythagorean theorem, based on the proportionality of similar triangles. Given a right-angled triangle with sides a, b, and hypotenuse c, we divide it into two smaller right-angled triangles by the perpendicular to the hypotenuse through the opposite vertex (the dashed line in Figure 2.18).
2.8 The Pythagorean theorem revisited
39
C
β
α a
b
A
β
α c1
D
c2
B
c Figure 2.18: Subdividing a right-angled triangle into similar triangles All three triangles are similar because they have the same angles α and β . If we look first at the angle α at A and the angle β at B, then
α +β =
π 2
because the angle sum of triangle ABC is π and the angle at C is π /2. But then it follows that angle ACD = β in triangle ACD (to make its angle sum = π ) and angle DCB = α in triangle DCB (to make its angle sum = π ). Now we use the proportionality of these triangles, calling the side opposite α in each triangle “short” and the side opposite β “long” for convenience. Comparing triangle ABC with triangle ADC, we get b c1 long side = = , hypotenuse c b
hence b2 = cc1 .
Comparing triangle ABC with triangle DCB, we get a c2 short side = = , hypotenuse c a
hence a2 = cc2 .
Adding the values of a2 and b2 just obtained, we finally get a2 + b2 = cc2 + cc1 = c(c1 + c2 ) = c2 and this is the Pythagorean theorem.
because c1 + c2 = c,
40
2 Euclid’s approach to geometry
This second proof is not really shorter than Euclid’s first (given in Section 2.5) when one takes into account the work needed to prove the proportionality of similar triangles. However, we often need similar triangles, so they are a standard tool, and a proof that uses standard tools is generally preferable to one that uses special machinery. Moreover, the splitting of a right-angled triangle into similar triangles is itself a useful tool—it enables us to construct the square root of any line segment.
Straightedge and compass construction of square roots Given any line segment l, construct the semicircle with diameter l + 1, and the perpendicular to the diameter where the segments √ 1 and l meet (Figure 2.19). Then the length h of this perpendicular is l.
h l
1
Figure 2.19: Construction of the square root To see why, construct the right-angled triangle with hypotenuse l + 1 and third vertex where the perpendicular meets the semicircle. We know that the perpendicular splits this triangle into two similar, and hence proportional, triangles. In the triangle on the left, l long side = . short side h In the triangle on the right, h long side = . short side 1 Because these ratios are equal by proportionality of the triangles, we have
hence h2 = l; that is, h =
√ l.
h l = , h 1
2.8 The Pythagorean theorem revisited
41
This result complements the constructions for the rational operations √ +, −, ×, and ÷ we gave in Chapter 1. The constructibility of these and was first pointed√out by Descartes in his book G´eom´etrie of 1637. Rational operations and are in fact precisely what can be done with straightedge and compass. When we introduce coordinates in Chapter 3 we will see that any “constructible point” has coordinates obtainable from the unit length 1 √ by +, −, ×, ÷, and .
Exercises
√ Now that we know how to construct the +, −, ×, ÷, and of given lengths, we can use algebra as a shortcut to decide whether certain figures are constructible by straightedge and √ compass. If we know that a certain figure is constructible from the length (1+ 5)/2, for example, √ then we know that the figure is constructible— 5)/2 is built from the unit length by the operaperiod—because the length (1 + √ tions +, ×, ÷, and . This is precisely the case for the regular pentagon, which was constructed by Euclid in Book IV, Proposition 11, using virtually all of the geometry he had developed up to that point. We also need nearly everything we have developed up to this point, but it fills less space than four books of the Elements! The following exercises refer to the regular pentagon of side 1 shown in Figure 2.20 and its diagonals of length x.
1 x
Figure 2.20: The regular pentagon 2.8.1 Use the symmetry of the regular pentagon to find similar triangles implying 1 x = , 1 x−1 that is, x2 − x − 1 = 0. 2.8.2 By finding the positive root √ of this quadratic equation, show that each diagonal has length x = (1 + 5)/2. 2.8.3 Now show that the regular pentagon is constructible.
42
2 Euclid’s approach to geometry
2.9 Discussion Euclid found the most important axiom of geometry—the parallel axiom— and he also identified the basic theorems and traced the logical connections between them. However, his approach misses certain fine points and is not logically complete. For example, in his very first proof (the construction of the equilateral triangle), he assumes that certain circles have a point in common, but none of his axioms guarantee the existence of such a point. There are many such situations, in which Euclid assumes something is true because it looks true in the diagram. Euclid’s theory of area is a whole section of his geometry that seems to have no geometric support. Its concepts seem more like arithmetic— addition, subtraction, and proportion—but its concept of multiplication is not the usual one, because multiplication of more than three lengths is not allowed. These gaps in Euclid’s approach to geometry were first noticed in the 19th century, and the task of filling them was completed by David Hilbert in his Grundlagen der Geometrie (Foundations of Geometry) of 1899. On the one hand, Hilbert introduced axioms of incidence and order, giving the conditions under which lines (and circles) meet. These justify the belief that “geometric objects behave as the pictures suggest.” On the other hand, Hilbert replaced Euclid’s theory of area with a genuine arithmetic, which he called segment arithmetic. He defined the sum and product of segments as we did in Section 1.4 and proved that these operations on segments have the same properties as ordinary sum and product. For example, a + b = b + a,
ab = ba,
a(b + c) = ab + ac,
and so on.
In the process, Hilbert discovered that the Pappus and Desargues theorems (Exercises 1.4.3 and 1.4.4) play a decisive role. The downside of Hilbert’s completion of Euclid is that it is lengthy and difficult. Nearly 20 axioms are required, and some key theorems are hard to prove. To some extent, this hardship occurs because Hilbert insists on geometric definitions of + and ×. He wants numbers to come from “inside” geometry rather than from “outside”. Thus, to prove that ab = ba he needs the theorem of Pappus, and to prove that a(bc) = (ab)c he needs the theorem of Desargues. Even today, the construction of segment arithmetic is an admirable feat. As Hilbert pointed out, it shows that Euclid was right to believe that the
2.9 Discussion
43
theory of proportion could be developed without new geometric axioms. Still, it is somewhat quixotic to build numbers “inside” Euclid’s geometry when they are brought from “outside” into nearly every other branch of geometry. It is generally easier to build geometry on numbers than the reverse, and Euclidean geometry is no exception, as I hope to show in Chapters 3 and 4. This is one reason for bypassing Hilbert’s approach, so I will merely list his axioms here. They are thoroughly investigated in Hartshorne’s Geometry: Euclid and Beyond or Hilbert’s own book, which is available in English translation. Hartshorne’s book has the clearest available derivation of ordinary geometry and segment arithmetic from the Hilbert axioms, so it should be consulted by anyone who wants to see Euclid’s approach taken to its logical conclusion. There is another reason to bypass Hilbert’s axioms, apart from their difficulty. In my opinion, Hilbert’s greatest geometric achievement was to build arithmetic, not in Euclidean geometry, but in projective geometry. As just mentioned, Hilbert found that the keys to segment arithmetic are the Pappus and Desargues theorems. These two theorems do not involve the concept of length, and so they really belong to a more primitive kind of geometry. This primitive geometry (projective geometry) has only a handful of axioms—fewer than the usual axioms for arithmetic—so it is more interesting to build arithmetic inside it. It is also less trouble, because we do not have to prove the Pappus and Desargues theorems. We will explain how projective geometry contains arithmetic in Chapters 5 and 6.
Hilbert’s axioms The axioms concern undefined objects called “points” and “lines,” the related concepts of “line segment,” “ray,” and “angle,” and the relations of “betweenness” and “congruence.” Following Hartshorne, we simplify Hilbert’s axioms slightly by stating some of them in a stronger form than necessary. The first group of axioms is about incidence: conditions for points to lie on lines or for lines to pass through points. I1. For any two points A, B, a unique line passes through A, B. I2. Every line contains at least two points. I3. There exist three points not all on the same line.
44
2 Euclid’s approach to geometry
I4. For each line L and point P not on L there is a unique line through P not meeting L (parallel axiom). The next group is about betweenness or order: a concept overlooked by Euclid, probably because it is too “obvious.” The first to draw attention to betweenness was the German mathematician Moritz Pasch, in the 1880s. We write A ∗ B ∗C to denote that B is between A and C. B1. If A ∗ B ∗C, then A, B,C are three points on a line and C ∗ B ∗ A. B2. For any two points A and B, there is a point C with A ∗ B ∗C. B3. Of three points on a line, exactly one is between the other two. B4. Suppose A, B,C are three points not in a line and that L is a line not passing through any of A, B,C. If L contains a point D between A and B, then L contains either a point between A and C or a point between B and C, but not both (Pasch’s axiom). The next group is about congruence of line segments and congruence of angles, both denoted by ∼ =. Thus, AB ∼ = CD means that AB and CD ∼ have equal length and ∠ABC = ∠DEF means that ∠ABC and ∠DEF are equal angles. Notice that C2 and C5 contain versions of Euclid’s Common Notion 1: “Things equal to the same thing are equal to each other.” C1. For any line segment AB, and any ray R originating at a point C, there is a unique point D on R with AB ∼ = CD. C2. If AB ∼ = CD and AB ∼ = EF, then CD ∼ = EF. For any AB, AB ∼ = AB. C3. Suppose A ∗ B ∗ C and D ∗ E ∗ F. If AB ∼ = DE and BC ∼ = EF, then AC ∼ DF. (Addition of lengths is well-defined.) = −→
−→
C4. For any angle ∠BAC, and any ray DF, there is a unique ray DE on a −→ ∼ ∠EDF. given side of DF with ∠BAC = C5. For any angles α , β , γ , if α ∼ = β and α ∼ = γ , then β ∼ = γ . Also, α ∼ = α. ∼ DE, AC = ∼ DF, C6. Suppose that ABC and DEF are triangles with AB = ∼ and ∠BAC = ∠EDF. Then, the two triangles are congruent, namely BC ∼ = EF, ∠ABC ∼ = ∠DEF, and ∠ACB ∼ = ∠DFE. (This is SAS.)
2.9 Discussion
45
Then there is an axiom about the intersection of circles. It involves the concept of points inside the circle, which are those points whose distance from the center is less than the radius. E. Two circles meet if one of them contains points both inside and outside the other. Next there is the so-called Archimedean axiom, which says that no length can be “infinitely large” relative to another. A. For any line segments AB and CD, there is a natural number n such that n copies of AB are together greater than CD. Finally, there is the so-called Dedekind axiom, which says that the line is complete, or has no gaps. It implies that its points correspond to real numbers. Hilbert wanted an axiom like this to force the plane of Euclidean geometry to be the same as the plane R2 of pairs of real numbers. D. Suppose the points of a line L are divided into two nonempty subsets A and B in such a way that no point of A is between two points of B and no point of B is between two points of A . Then, a unique point P, either in A or B, lies between any other two points, of which one is in A and the other is in B. Axiom D is not needed to derive any of Euclid’s theorems. They do not involve all real numbers but only the so-called constructible numbers originating from straightedge and compass constructions. However, who can be sure that we will never need nonconstructible points? One of the most important numbers in geometry, π , is nonconstructible! (Because the circle cannot be squared.) Thus, it seems prudent to use Axiom D so that the line is complete from the beginning. In Chapter 3, we will take the real numbers as the starting point of geometry, and see what advantages this may have over the Euclid–Hilbert approach. One clear advantage is access to algebra, which reduces many geometric problems to simple calculations. Algebra also offers some conceptual advantages, as we will see.
3 Coordinates P REVIEW Around 1630, Pierre de Fermat and Ren´e Descartes independently discovered the advantages of numbers in geometry, as coordinates. Descartes was the first to publish a detailed account, in his book G´eom´etrie of 1637. For this reason, he gets most of the credit for the idea and the coordinate approach to geometry became known as Cartesian (from the old way of writing his name: Des Cartes). Descartes thought that geometry was as Euclid described it, and that numbers merely assist in studying geometric figures. But later mathematicians discovered objects with “non-Euclidean” properties, such as “lines” having more than one “parallel” through a given point. To clarify this situation, it became desirable to define points, lines, length, and so on, and to prove that they satisfy Euclid’s axioms. This program, carried out with the help of coordinates, is called the arithmetization of geometry. In the first three sections of this chapter, we do the main steps, using the set R of real numbers to define the Euclidean plane R2 and the points, lines, and circles in it. We also define the concepts of distance and (briefly) angle, and show how some crucial axioms and theorems follow. However, arithmetization does much more. • It gives an algebraic description of constructibility by straightedge and compass (Section 3.4), which makes it possible to prove that certain figures are not constructible. • It enables us to define what it means to “move” a geometric figure (Section 3.6), which provides justification for Euclid’s proof of SAS, and raises a new kind of geometric question (Section 3.7): What kinds of “motion” exist?
46
3.1 The number line and the number plane
47
3.1 The number line and the number plane
y-axis
The set R of real numbers results from filling the √ gaps in the set Q of rational numbers with irrational numbers, such as 2. This innovation enables us to consider R as a line, because it has no gaps and the numbers in it are ordered just as we imagine points on a line to be. We say that R, together with its ordering, is a model of the line. One of our goals in this chapter is to use R to build a model for all of Euclidean plane geometry: a structure containing “lines,” “circles,” “line segments,” and so on, with all of the properties required by Euclid’s or Hilbert’s axioms. The first step is to build the “plane,” and in this we are guided by the properties of parallels in Euclid’s geometry. We imagine a pair of perpendicular lines, called the x-axis and the y-axis, intersecting at a point O called the origin (Figure 3.1). We interpret the axes as number lines, with O the number 0 on each, and we assume that the positive direction on the x-axis is to the right and that the positive direction on the y-axis is upward.
P = (a, b)
b
x-axis O
a Figure 3.1: Axes and coordinates
Through any point P, there is (by the parallel axiom) a unique line parallel to the y-axis and a unique line parallel to the x-axis. These two lines meet the x-axis and y-axis at numbers a and b called the x- and ycoordinates of P, respectively. It is important to remember which number is on the x-axis and which is on the y-axis, because obviously the point with x-coordinate = 3 and y-coordinate = 4 is different from the point with x-coordinate = 4 and y-coordinate = 3 (just as the intersection of 3rd Street and 4th Avenue is different from the intersection of 4th Street and 3rd Avenue).
48
3
Coordinates
To keep the x-coordinate a and the y-coordinate b in their places, we use the ordered pair (a, b). For example, (3, 4) is the point with x-coordinate = 3 and y-coordinate = 4, whereas (4, 3) is the point with x-coordinate = 4 and y-coordinate = 3. The ordered pair (a, b) specifies P uniquely because any other point will have at least one different parallel passing through it and hence will differ from P in either the x- or y-coordinate. Thus, given the existence of a number line R whose points are real numbers, we also have a number plane whose points are ordered pairs of real numbers. We often write this number plane as R × R or R 2 .
3.2 Lines and their equations As mentioned in Chapter 2, one of the most important consequences of the parallel axiom is the Thales theorem and hence the proportionality of similar triangles. When coordinates are introduced, this allows us to define the property of straight lines known as slope. You know from high-school mathematics that slope is the quotient “rise over run” and, more importantly, that the value of the slope does not depend on which two points of the line define the rise and the run. Figure 3.2 shows why. B β
B β A
α C
A
α
Figure 3.2: Why the slope of a line is constant In this figure, we have two segments of the same line: • AB, for which the rise is |BC| and the run is |AC|, and • A B , for which the rise is |BC | and the run is |AC |.
C
3.2 Lines and their equations
49
The angles marked α are equal because AC and AC are parallel, and the angles marked β are equal because the BC and BC are parallel. Also, the angles at C and C are both right angles. Thus, triangles ABC and A BC are similar, and so their corresponding sides are proportional. In particular, |BC| |BC | = , |AC| |AC | that is, slope = constant. Now suppose we are given a line of slope a that crosses the y-axis at the point Q where y = c (Figure 3.3). If P = (x, y) is any point on this line, then the rise from Q to P is y − c and the run is x. Hence y−c x and therefore, multiplying both sides by x, y − c = ax, that is, slope = a =
y = ax + c. This equation is satisfied by all points on the line, and only by them, so we call it the equation of the line. y (x, y) = P y−c (0, c) = Q
x x
O Figure 3.3: Typical point on the line
Almost all lines have equations of this form; the only exceptions are lines that do not cross the y-axis. These are the vertical lines, which also do not have a slope as we have defined it, although we could say they have infinite slope. Such a line has an equation of the form x = c,
for some constant c.
50
3
Coordinates
Thus, all lines have equations of the form ax + by + c = 0,
for some constants a, b, and c,
called a linear equation in the variables x and y. Up to this point we have been following the steps of Descartes, who viewed equations of lines as information deduced from Euclid’s axioms (in particular, from the parallel axiom). It is true that Euclid’s axioms prompt us to describe lines by linear equations, but we can also take the opposite view: Equations define what lines and curves are, and they provide a model of Euclid’s axioms—showing that geometry follows from properties of the real numbers. In particular, if a line is defined to be the set of points (x, y) in the number plane satisfying a linear equation then we can prove the following statements that Euclid took as axioms: • there is a unique line through any two distinct points, • for any line L and point P outside L , there is a unique line through P not meeting L . Because these statements are easy to prove, we leave them to the exercises.
Exercises Given distinct points P1 = (x1 , y1 ) and P2 = (x2 , y2 ), suppose that P = (x, y) is any point on a line through P1 and P2 . 3.2.1 By equating slopes, show that x and y satisfy the equation y2 − y1 y − y1 = if x2 = x1 . x2 − x1 x − x1 3.2.2 Explain why the equation found in Exercise 3.2.1 is the equation of a straight line. 3.2.3 What happens if x2 = x1 ? Parallel lines, not surprisingly, turn out to be lines with the same slope. 3.2.4 Show that distinct lines y = ax + c and y = a x + c have a common point unless they have the same slope (a = a ). Show that this is also the case when one line has infinite slope. 3.2.5 Deduce from Exercise 3.2.4 that the parallel to a line L is the unique line through P with the same slope as L . 3.2.6 If L has equation y = 3x, what is the equation of the parallel to L through P = (2, 2)?
3.3 Distance
51
3.3 Distance We introduce the concept of distance or length into the number plane R 2 much as we introduce lines. First we see what Euclid’s geometry suggests distance should mean; then we turn around and take the suggested meaning as a definition. Suppose that P1 = (x1 , y1 ) and P2 = (x2 , y2 ) are any two points in R2 . Then it follows from the meaning of coordinates that there is a right-angled triangle as shown in Figure 3.4, and that |P1 P2 | is the length of its hypotenuse. y P2 = (x2 , y2 ) y2 − y1 (x1 , y1 ) = P1
x2 − x1
x
O Figure 3.4: The triangle that defines distance
The vertical side of the triangle has length y2 − y1 , and the horizontal side has length x2 − x1 . Then it follows from the Pythagorean theorem that |P1 P2 |2 = (x2 − x1 )2 + (y2 − y1 )2 , and therefore, |P1 P2 | =
(x2 − x1 )2 + (y2 − y1 )2 .
(*)
Thus, it is sensible to define the distance |P1 P2 | between any two points P1 and P2 by the formula (*). If we do this, the Pythagorean theorem is virtually “true by definition.” It is certainly true when the right-angled triangle has a vertical side and a horizontal side, as in Figure 3.4. And we will see later how to rotate any right-angled triangle to such a position (without changing the lengths of its sides).
52
3
Coordinates
The equation of a circle The distance formula (*) leads immediately to the equation of a circle, as follows. Suppose we have a circle with radius r and center at the point P = (a, b). Then any point Q = (x, y) on the circle is at distance r from P, and hence formula (*) gives: r = |PQ| = (x − a)2 + (y − b)2 . Squaring both sides, we get (x − a)2 + (y − b)2 = r2 . We call this the equation of the circle because it is satisfied by any point (x, y) on the circle, and only by such points.
The equidistant line of two points A circle is the set of points equidistant from a point—its center. It is also natural to ask: What is the set of points equidistant from two points in R 2 ? Answer: The set of points equidistant from two points is a line. To see why, let the two points be P1 = (a1 , b1 ) and P2 = (a2 , b2 ). Then a point P = (x, y) is equidistant from P1 and P2 if |PP1 | = |PP2 |, that is, if x and y satisfy the equation 2 2 (x − a1 ) + (y − b1 ) = (x − a2 )2 + (y − b2 )2 . Squaring both sides of this equation, we get (x − a1 )2 + (y − b1 )2 = (x − a2 )2 + (y − b2 )2 . Expanding the squares gives x2 − 2a1 x + a21 + y2 − 2b1 y + b21 = x2 − 2a2 x + a22 + y2 − 2b2 y + b22 . The important thing is that the x2 and y2 terms now cancel, which leaves the linear equation 2(a2 − a1 )x + 2(b2 − b1 )y + (b21 − b22 ) = 0. Thus, the points P = (x, y) equidistant from P1 and P2 form a line.
3.4 Intersections of lines and circles
53
Exercises An interesting application of equidistant lines is the following. 3.3.1 Show that any three points not in a line lie on a unique circle. (Hint: the center of the circle is equidistant from the three points.) The equations of lines and circles enable us to prove many geometric theorems by algebra, as Descartes realized. In fact, they greatly expand the scope of geometry by allowing many curves to be described by equations. But algebra is also useful in proving that certain quantities are not equal. One example is the triangle inequality. 3.3.2 Consider any triangle, which for convenience we take to have one vertex at O = (0, 0), one at P = (x1 , 0) with x1 > 0, and one at Q = (x2 , y2 ). Show that |OP| = x1 , |PQ| = (x2 − x1 )2 + y22 , |OQ| = x22 + y22 . The triangle inequality states that |OP| + |PQ| > |OQ| (any two sides of a triangle are together greater than the third). To prove this statement, it suffices to show that (|OP| + |PQ|)2 > |OQ|2 . 3.3.3 Show that (|OP| + |PQ|)2 − |OQ|2 = 2x1 (x2 − x1 )2 + y22 − (x2 − x1 ) . 3.3.4 Show that the term in square brackets in Exercise 3.3.3 is positive if y2 = 0, and hence that the triangle inequality holds in this case. 3.3.5 If y2 = 0, why is this not a problem? Later we will give a more sophisticated approach to the triangle inequality, which does not depend on choosing a special position for the triangle.
3.4 Intersections of lines and circles Now that lines and circles are defined by equations, we can give exact algebraic equivalents of straightedge and compass operations: • Drawing a line through given points corresponds to finding the equation of the line through given points (x1 , y1 ) and (x2 , y2 ). The slope y−y1 1 between these two points is yx22 −y −x1 , which must equal the slope x−x1 between the general point (x, y) and the special point (x1 , y1 ), so the equation is y − y1 y2 − y1 = . x − x1 x2 − x1
54
3
Coordinates
Multiplying both sides by (x − x1 )(x2 − x1 ), we get the equivalent equation (y − y1 )(x2 − x1 ) = (x − x1 )(y2 − y1 ), or (y2 − y1 )x − (x2 − x1 )y − x1 y2 + y1 x2 = 0. • Drawing a circle with given center and radius corresponds to finding the equation of the circle with given center (a, b) and given radius r, which is (x − a)2 + (y − b)2 = r2 . • Finding new points as intersections of previously drawn lines and circles corresponds to finding the solution points of – a pair of equations of lines, – a pair of equations of circles, – the equation of a line and the equation of a circle. For example, to find the intersection of the two circles (x − a1 )2 + (y − b1 )2 = r12 and
(x − a2 )2 + (y − b2 )2 = r22 ,
we expand the equations of the circles as x2 − 2a1 x + a21 + y2 − 2b1 y + b21 − r12 = 0, x
2
− 2a2 x + a22 + y2 − 2b2 y + b22 − r22
= 0,
(1) (2)
and subtract Equation (2) from Equation (1). The x2 and y2 terms cancel, and we are left with the linear equation in x and y: 2(a2 − a1 )x + 2(b2 − b2 )y + r22 − r12 = 0.
(3)
We can solve Equation (3) for either x or y. Then substituting the solution of (3) in (1) gives a quadratic equation for either y or x. If the equation is of the form Ax2 + Bx +C = 0, then we know that the solutions are √ −B ± B2 − 4AC x= . 2A
3.5 Angle and slope
55
Solving linear equations requires only the √ operations +, −, ×, and ÷, and the quadratic formula shows that is the only additional operation needed to solve quadratic equations. Thus, all intersection points involved in a straightedge and compass √ construction can be found with the operations +, −, ×, ÷, and . √ Now recall from Chapters 1 and 2 that the operations +, −, ×, ÷, and can be carried out by straightedge and compass. Hence, we get the following result: Algebraic criterion for constructibility. A point is constructible (starting from the points 0 and 1) if and only if its coordinates are obtainable from √ the number 1 by the operations +, −, ×, ÷, and . The algebraic criterion for constructibility was discovered by Descartes, and its greatest virtue is that it enables us to prove that certain figures or points are not constructible. For example, one can prove that the number √ 3 2 is not constructible by showing that it cannot be expressed by a finite number of square roots, and one can prove that the angle π /3 cannot be trisected by showing that cos π9 also cannot be expressed by a finite number of square roots. These results were not proved until the 19th century, by Pierre Wantzel. Rather sophisticated algebra is required, because one has to go beyond Descartes’ concept of constructibility to survey the totality of constructible numbers.
Exercises 3.4.1 Find the intersections of the circles x2 + y2 = 1 and (x − 1)2 + (y − 2)2 = 4. 3.4.2 Check the plausibility of your answer to Exercise 3.4.1 by a sketch of the two circles. 3.4.3 The line x + 2y − 1 = 0 found by eliminating the x2 and y2 from the equations of the circles should have some geometric meaning. What is it?
3.5 Angle and slope The concept of distance is easy to handle in coordinate geometry because the distance between points (x1 , y1 ) and (x2 , y2 ) is an algebraic function of their coordinates, namely (x2 − x1 )2 + (y2 − y1 )2 .
56
3
Coordinates
This is not the case for the concept of angle. The angle θ between a line y = tx and the x-axis is tan−1 t, and the function tan−1 t is not an algebraic function. Nor is its inverse function t = tan θ or the related functions sin θ (sine) and cos θ (cosine). To stay within the world of algebra, we have to work with the slope t rather than the angle θ . Lines make the same angle with the x-axis if they have the same slope, but to test equality of angles in general we need the concept of relative slope: If line L1 has slope t1 and line L2 has slope t2 , then the slope of L1 relative to L2 is defined to be t1 − t 2 . ± 1 + t 1 t2 This awkward definition comes from the formula you have probably seen in trigonometry, tan(θ1 − θ2 ) =
tan θ1 − tan θ2 , 1 + tan θ1 tan θ2
by taking t1 = tan θ1 and t2 = tan θ2 . The reason for the ± sign and the absolute value is that the slopes t1 , t2 alone do not specify an angle—they specify only a pair of lines and hence a pair of angles that add to a straight angle. (For more on using relative slope to discuss equality of angles, see Hartshorne’s Geometry: Euclid and Beyond, particularly pp. 141–155.) At any rate, with some care it is possible to use the concept of relative slope to test algebraically whether angles are equal. The concept also makes it possible to state the SAS and ASA axioms in coordinate geometry, and to verify that all of Euclid’s and Hilbert’s axioms hold. We omit the details because they are laborious, and because we can approach SAS and ASA differently now that we have coordinates. Specifically, it becomes possible to define the concept of “motion” that Euclid appealed to in his proof of SAS! This will be done in the next section.
Exercises The most useful instance of relative slope is where the lines are perpendicular. 3.5.1 Show that lines of slopes t1 and t2 are perpendicular just in case t1t2 = −1. 3.5.2 Use the condition for perpendicularity found in Exercise 3.5.1 to show that the line from (1, 0) to (3, 4) is perpendicular to the line from (0, 2) to (4, 0).
3.6 Isometries
57
In the next section, we will define a rotation about O to be a transformation rc,s of R2 depending on two real numbers c and s such that c2 + s2 = 1. The transformation rc,s sends the point (x, y) to the point (cx − sy, sx + cy). It will be explained in the next section why it is reasonable to call this a “rotation about O,” and why c = cos θ and s = sin θ , where θ is the angle of rotation. For the moment, suppose that this is the case, and consider the effect of two rotations rc1 ,s1 and rc2 ,s2 , where c1 = cos θ1 ,
s1 = sin θ1 ;
c2 = cos θ2 ,
s2 = sin θ2 .
This thought experiment leads us to proofs of the formulas for cos, sin, and tan of θ1 + θ 2 : 3.5.3 Show that the outcome of rc1 ,s1 and rc2 ,s2 is to send (x, y) to ((c1 c2 − s1 s2 )x − (s1 c2 + c1 s2 )y, (s1 c2 + c1 s2 )x + (c1 c2 − s1 s2 )y) . 3.5.4 Assuming that rc1 ,s1 really is a rotation about O through angle θ1 , and rc2 ,s2 really is a rotation about O through angle θ2 , deduce from Exercise 3.5.3 that cos(θ1 + θ2 ) = cos θ1 cos θ2 − sin θ1 sin θ2 , sin(θ1 + θ2 ) = sin θ1 cos θ2 + cos θ1 sin θ2 . 3.5.5 Deduce from Exercise 3.5.4 that tan(θ1 + θ2 ) =
tan θ1 + tan θ2 , 1 − tan θ1 tan θ2
tan(θ1 − θ2 ) =
tan θ1 − tan θ2 . 1 + tan θ1 tan θ2
hence
3.6 Isometries A possible weakness of our model of the plane is that it seems to single out a particular point (the origin O) and particular lines (the x- and y-axes). In Euclid’s plane, each point is like any other point and each line is like any other line. We can overcome the apparent bias of R 2 by considering transformations that allow any point to become the origin and any line to become the x-axis. As a bonus, this idea gives meaning to the idea of “motion” that Euclid tried to use in his attempt to prove SAS. A transformation of the plane is simply a function f : R2 → R2 , in other words, a function that sends points to points.
58
3
Coordinates
A transformation f is called an isometry (from the Greek for “same length”) if it sends any two points, P1 and P2 , to points f (P1 ) and f (P2 ) the same distance apart. Thus, an isometry is a function f with the property | f (P1 ) f (P2 )| = |P1 P2 | for any two points P1 , P2 . Intuitively speaking, an isometry “moves the plane rigidly” because it preserves the distance between points. There are many isometries of the plane, but they can be divided into a few simple and obvious types. We show examples of each type below, and, in the next section, we explain why only these types exist. You will notice that certain isometries (translations and rotations) make it possible to move the origin to any point in the plane and the x-axis to any line. Thus, R2 is really like Euclid’s plane, in the sense that each point is like any other point and each line is like any other line. This property entitles us to choose axes wherever it is convenient. For example, we are entitled to prove the triangle inequality, as suggested in the Exercises to Section 3.3, by choosing one vertex of the triangle at O and another on the positive x-axis. Translations A translation moves each point of the plane the same distance in the same direction. Each translation depends on two constants a and b, so we denote it by ta,b . It sends each point (x, y) to the point (x + a, y + b). It is obvious that a translation preserves the distance between any two points, but it is worth checking this formally—so as to know what to do in less obvious cases. So let P1 = (x1 , y1 ) and P2 = (x2 , y2 ). It follows that ta,b (P1 ) = (x1 + a, y1 + b),
ta,b (P2 ) = (x2 + a, y2 + b)
and therefore, (x2 + a − x1 − a)2 + (y2 + b − y1 − b)2 = (x2 − x1 )2 + (y2 − y1 )2
|ta,b (P1 )ta,b (P2 )| =
= |P1 P2 |,
as required.
3.6 Isometries
59
Rotations We think of a rotation as something involving an angle θ , but, as mentioned in the previous section, it is more convenient to work algebraically with cos θ and sin θ . These are simply two numbers c and s such that c 2 +s2 = 1, so we will denote a rotation of the plane about the origin by r c,s . The rotation rc,s sends the point (x, y) to the point (cx − sy, sx + cy). It is not obvious why this transformation should be called a rotation, but it becomes clearer after we check that rc,s preserves lengths. If we let P1 = (x1 , y1 ) and P2 = (x2 , y2 ) again, it follows that rc,s (P1 ) = (cx1 − sy1 , sx1 + cy1 ), and therefore, |rc,s (P1 )rc,s (P2 )| =
= = =
rc,s (P2 ) = (cx2 − sy2 , sx2 + cy2 )
[c(x2 − x1 ) − s(y2 − y1 )]2 + [s(x2 − x1 ) + c(y2 − y1 )]2 c2 (x2 − x1 )2 − 2cs(x2 − x1 )(y2 − y1 ) + s2 (y2 − y1 )2 +s2 (x2 − x1 )2 + 2cs(x2 − x1 )(y2 − y1 ) + c2 (y2 − y1 )2 (c2 + s2 )(x2 − x1 )2 + (c2 + s2 )(y2 − y1 )2 (x2 − x1 )2 + (y2 − y1 )2
because c2 + s2 = 1
= |P1 P2 |. Thus, rc,s preserves lengths. Also, rc,s sends (0, 0) to itself, and it moves (1, 0) to (c, s) and (0, 1) to (−s, c), which is exactly what rotation about O through angle θ does (see Figure 3.5). We will see in the next section that only one isometry of the plane moves these three points in this manner. (0, 1)
(−s, c)
(c, s) = (cos θ , sin θ )
O
(1, 0)
Figure 3.5: Movement of points by a rotation
60
3
Coordinates
Reflections The easiest reflection to describe is reflection in the x-axis, which sends P = (x, y) to P = (x, −y). Again it is obvious that this is an isometry, but we can check by calculating the distance between reflected points P1 and P2 (Exercise 3.6.1). We can reflect the plane in any line, and we can do this by combining reflection in the x-axis with translations and rotations. For example, reflection in the line y = 1 (which is parallel to the x-axis) is the result of the following three isometries: • t0,−1 , a translation that moves the line y = 1 to the x-axis, • reflection in the x-axis, • t0,1 , which moves the x-axis back to the line y = 1. In general, we can do a reflection in any line L by moving L to the x-axis by some combination of translation and rotation, reflecting in the x-axis, and then moving the x-axis back to L . Reflections are the most fundamental isometries, because any isometry is a combination of them, as we will see in the next section. In particular, any translation is a combination of two reflections, and any rotation is a combination of two reflections (see Exercises 3.6.2–3.6.4). Glide reflections A glide reflection is the result of a reflection followed by a translation in the direction of the line of reflection. For example, if we reflect in the x-axis, sending (x, y) to (x, −y), and follow this with the translation t 1,0 of length 1 in the x-direction, then (x, y) ends up at (x + 1, −y). A glide reflection with nonzero translation length is different from the three types of isometry previously considered. • It is not a translation, because a translation maps any line in the direction of translation into itself, whereas a glide reflection maps only one line into itself (namely, the line of reflection). • It is not a rotation, because a rotation has a fixed point and a glide reflection does not. • It is not a reflection, because a reflection also has fixed points (all points on the line of reflection).
3.7 The three reflections theorem
61
Exercises 3.6.1 Check that reflection in the x-axis preserves the distance between any two points. When we combine reflections in two lines, the nature of the outcome depends on whether the lines are parallel. 3.6.2 Reflect the plane in the x-axis, and then in the line y = 1/2. Show that the resulting isometry sends (x, y) to (x, y + 1), so it is the translation t0,1 . 3.6.3 Generalize the idea of Exercise 3.6.2 to show that the combination of reflections in parallel lines, distance d/2 apart, is a translation through distance d, in the direction perpendicular to the lines of reflection. 3.6.4 Show, by a suitable picture, that the combination of reflections in lines that meet at angle θ /2 is a rotation through angle θ , about the point of intersection of the lines. Another way to put the result of Exercise 3.6.4 is as follows: Reflections in any two lines meeting at the same angle θ /2 at the same point P give the same outcome. This observation is important for the next three exercises (where pictures will also be helpful). 3.6.5 Show that reflections in lines L , M , and N (in that order) have the same outcome as reflections in lines L , M , and N , where M is perpendicular to N . 3.6.6 Next show that reflections in lines L , M , and N have the same outcome as reflections in lines L , M , and N , where M is parallel to L and N is perpendicular to M . 3.6.7 Deduce from Exercise 3.6.6 that the combination of any three reflections is a glide reflection.
3.7 The three reflections theorem We saw in Section 3.3 that the points equidistant from two points A and B form a line, which implies that isometries of the plane are very simple: An isometry f of R2 is determined by the images f (A), f (B), f (C) of three points A, B,C not in a line. The proof follows from three simple observations: • Any point P in R2 is determined by its distances from A, B,C. Because if Q is another point with the same distances from A, B,C as P, then A, B,C lie in the equidistant line of P and Q, contrary to the assumption that A, B,C are not in a line.
62
3
Coordinates
• The isometry f preserves distances (by definition of isometry), so f (P) lies at the same respective distances from f (A), f (B), f (C) as P does from A, B,C. • There is only one point at given distances from f (A), f (B), f (C) because these three points are not in a line—in fact they form a triangle congruent to triangle ABC, because f preserves distances. Thus, the image f (P) of any point P—and hence the whole isometry f —is determined by the images of three points A, B,C not in a line. This “three point determination theorem” gives us the: Three reflections theorem. Any isometry of R2 is a combination of one, two, or three reflections. Given an isometry f , we choose three points A, B,C not in a line, and we look for a combination of reflections that sends A to f (A), B to f (B), and C to f (C). Such a combination is necessarily equal to f . We can certainly send A to f (A) by reflection in the equidistant line of A and f (A). Call this reflection rA . Now rA sends B to rA (B), so if rA (B) = f (B) we need to do nothing more for B. If rA (B) = f (B), we can send rA (B) to f (B) by reflection rB in the equidistant line of rA (B) and f (B). Fortunately, f (A) = rA (A) lies on this line, because the distance from f (A) to f (B) equals the distance from r A (A) to rA (B) (because f and rA are isometries). Thus, rB does not move f (A), and the combination of rA followed by rB sends A to f (A) and B to f (B). The argument is similar for C. If C has already been sent to f (C), we are done. If not, we reflect in the line equidistant from f (C) and the point where C has been sent so far. It turns out (by a check of equal distances like that made for f (A) above) that f (A) and f (B) already lie on this line, so they are not moved. Thus, we finally have a combination of no more than three reflections that moves A to f (A), B to f (B), and C to f (C), as required. Now of course, one reflection is a reflection, and we found in the previous exercise set that combinations of two reflections are translations and rotations, and that combinations of three reflections are glide reflections (which include reflections). Thus, an isometry of R2 is either a translation, a rotation, or a glide reflection.
3.8 Discussion
63
Exercises Given three points A, B,C and the points f (A), f (B), f (C) to which they are sent by an isometry f , it is possible to find three reflections that combine to form f by following the steps in the proof above. However, if one merely wants to know what kind of isometry f is—translation, rotation, or glide reflection—then the answer can be found more simply. To fix ideas, we take the initial three points to be A = (0, 1), B = (0, 0), and C = (1, 0). You will probably find it helpful to sketch the triples of points f (A), f (B), f (C) given in the following exercises. 3.7.1 Suppose that f (A) = (1.4, 2), f (B) = (1.4, 1), and f (C) = (2.4, 1). Is f a translation or a rotation? How can you tell that f is not a glide reflection? 3.7.2 Suppose that f (A) = (0.4, 1.8), f (B) = (1, 1), and f (C) = (1.8, 1.6). We can tell that f is not a translation or glide reflection (hence, it must be a rotation). How? 3.7.3 Suppose that f (A) = (1.8, 1.6), f (B) = (1, 1), and f (C) = (0.4, 1.8). How do I know that this is a glide reflection? 3.7.4 State a simple test for telling whether f is a translation, rotation, or glide reflection from the positions of f (A), f (B), and f (C).
3.8 Discussion The discovery of coordinates is rightly considered a turning point in the development of mathematics because it reveals a vast new panorama of geometry, open to exploration in at least three different directions. • Description of curves by equations, and their analysis by algebra. This direction is called algebraic geometry, and the curves described by polynomial equations are called algebraic curves. Straight lines, described by the linear equations ax+by+c = 0, are called curves of degree 1. Circles, described by the equations (x−a)2 +(y−b)2 = r2 , are curves of degree 2, and so on. One can see that there are curves of arbitrarily high degree, so most of algebraic geometry is beyond the scope of this book. Even the curves of degree 3 are worth a book of their own, so for them, and other algebraic curves, we refer readers elsewhere. Two excellent books, which show how algebraic geometry relates to other parts of mathematics, are Elliptic Curves by H. P. McKean and V. Moll and Plane Algebraic Curves by E. Brieskorn and H. Kno¨ rrer.
64
3
Coordinates
• Algebraic study of objects described by linear equations (such as lines and planes). Even this is a big subject, called linear algebra. Although it is technically part of algebraic geometry, it has a special flavor, very close to that of Euclidean geometry. We explore plane geometry from the viewpoint of linear algebra in Chapter 4, and later we make some brief excursions into three and four dimensions. The real strength of linear algebra is its ability to describe spaces of any number of dimensions in geometric language. Again, this investigation is beyond our scope, but we will recommend additional reading at the appropriate places. • The study of transformations, which draws on the special branch of algebra known as group theory. Because many geometric transformations are described by linear equations, this study overlaps with linear algebra. The role of transformations was first emphasized by the German mathematician Felix Klein, in an address he delivered at the University of Erlangen in 1872. His address, known by its German name the Erlanger Programm, characterizes geometry as the study of transformation groups and their invariants. So far, we have seen only one transformation group and a handful of invariants—the group of isometries of R2 and what it leaves invariant (length, angle, straightness)—so the importance of Klein’s idea can hardly be clear yet. However, in Chapter 4 we introduce a very different group of transformations and a very different invariant—the projective transformations and the cross-ratio—so readers are asked to bear with us. In Chapters 7 and 8, we develop Klein’s idea in some generality and give another significant example, the geometry of the “non-Euclidean” plane.
4 Vectors and Euclidean spaces P REVIEW In this chapter, we process coordinates by linear algebra. We view points as vectors that can be added and multiplied by numbers, and we introduce the inner product of vectors, which gives an efficient algebraic method to deal with both lengths and angles. We revisit some theorems of Euclid to see where they fit in the world of vector geometry, and we become acquainted with some theorems that are particularly natural in this environment. For plane geometry, the appropriate vectors are ordered pairs (x, y) of real numbers. We add pairs according to the rule (u1 , u2 ) + (v1 , v2 ) = (u1 + v1 , u2 + v2 ), and multiply a pair by a real number a according to the rule a(u1 , u2 ) = (au1 , au2 ). These vector operations do not involve the concept of length or distance; yet they enable us to discuss certain ratios of lengths and to prove the theorems of Thales and Pappus. The concept of distance is introduced through the concept of inner product u · v of vectors u and v. If u = (u1 , u2 ) and v = (v1 , v2 ), then u · v = u1 v1 + u2 v2 . The inner product gives us distance because u · u = |u|2 , where |u| is the distance of u from the origin 0. It also gives us angle because u · v = |u||v| cos θ , where θ is the angle between the directions of u and v from 0.
65
66
4 Vectors and Euclidean spaces
4.1 Vectors Vectors are mathematical objects that can be added, and multiplied by numbers, subject to certain rules. The real numbers are the simplest example of vectors, and the rules for sums and multiples of any vectors are just the following properties of sums and multiples of numbers: u+v = v+u
1u = u
u + (v + w) = (u + v) + w
a(u + v) = au + av
u+0 = u
(a + b)u = au + bu
u + (−u) = 0
a(bu) = (ab)u.
These rules obviously hold when a, b, 1, u, v, w, 0 are all numbers, and 0 is the ordinary zero. They also hold when u, v, w are points in the plane R2 , if we interpret 0 as (0, 0), + as the vector sum defined for u = (u1 , u2 ) and v = (v1 , v2 ) by (u1 , u2 ) + (v1 + v2 ) = (u1 + v1 , u2 + v2 ), and au as the scalar multiple defined by a(u1 , u2 ) = (au1 , au2 ). The vector sum is geometrically interesting, because u + v is the fourth vertex of a parallelogram formed by the points 0, u, and v (Figure 4.1).
u + v = (u1 + v1 , u2 + v2 ) v = (v1 , v2 )
u = (u1 , u2 ) 0 Figure 4.1: The parallelogram rule for vector sum
4.1 Vectors
67
In fact, the rule for forming the sum of two vectors is often called the “parallelogram rule.” Scalar multiplication by a is also geometrically interesting, because it represents magnification by the factor a. It magnifies, or dilates, the whole plane by the factor a, transforming each figure into a similar copy of itself. Figure 4.2 shows an example of this with a = 2.5.
aw
au w av
u v 0
Figure 4.2: Scalar multiplication as a dilation of the plane
Real vector spaces It seems that the operations of vector addition and scalar multiplication capture some geometrically interesting features of a space. With this in mind, we define a real vector space to be a set V of objects, called vectors, with operations of vector addition and scalar multiplication satisfying the following conditions: • If u and v are in V , then so are u + v and au for any real number a. • There is a zero vector 0 such that u + 0 = u for each vector u. Each u in V has a additive inverse −u such that u + (−u) = 0. • Vector addition and scalar multiplication on V have the eight properties listed at the beginning of this section.
68
4 Vectors and Euclidean spaces
It turns out that real vector spaces are a natural setting for Euclidean geometry. We must introduce extra structure, which is called the inner product, before we can talk about length and angle. But once the inner product is there, we can prove all theorems of Euclidean geometry, often more efficiently than before. Also, we can uniformly extend geometry to any number of dimensions by considering the space Rn of ordered n-tuples of real numbers (x1 , x2 , . . . , xn ). For example, we can study three-dimensional Euclidean geometry in the space of ordered triples R3 = {(x1 , x2 , x3 ) : x1 , x2 , x3 ∈ R}, where the sum of u = (u1 , u2 , u3 ) and v = (v1 , v2 , v3 ) is defined by (u1 , u2 , u3 ) + (v1 , v2 , v3 ) = (u1 + v1 , u2 + v2 , u3 + v3 ) and the scalar multiple au is defined by a(u1 , u2 , u3 ) = (au1 , au2 , au3 ).
Exercises It is obvious that R2 has the eight properties of a real vector space. However, it is worth noting that R2 “inherits” these eight properties from the corresponding properties of real numbers. For example, the property u + v = v + u (called the commutative law) for vector addition is inherited from the corresponding commutative law for number addition, u + v = v + u, as follows: u + v = (u1 , u2 ) + (v1 + v2 ) = (u1 + v1 , u2 + v2 ) by definition of vector addition = (v1 + u1 , v2 + u2 ) by commutative law for numbers = (v1 , v2 ) + (u1 , u2 ) by definition of vector addition = v + u. 4.1.1 Check that the other seven properties of a vector space for R2 are inherited from corresponding properties of R. 4.1.2 Similarly check that Rn has the eight properties of a vector space. The term “dilation” for multiplication of all vectors in R2 (or Rn for that matter) by a real number a goes a little beyond the everyday meaning of the word in the case when a is smaller than 1 or negative. 4.1.3 What is the geometric meaning of the transformation of R2 when every vector is multiplied by −1? Is it a rotation? 4.1.4 Is it a rotation of R3 when every vector is multiplied by −1?
4.2 Direction and linear independence
69
4.2 Direction and linear independence Vectors give a concept of direction in R2 by representing lines through 0. If u is a nonzero vector, then the real multiples au of u make up the line through 0 and u, so we call them the points “in direction u from 0.” (You may prefer to say that −u is in the direction opposite to u, but it is simpler to associate direction with a whole line, rather than a half line.) Nonzero vectors u and v, therefore, have different directions from 0 if neither is a multiple of the other. It follows that such u and v are linearly independent; that is, there are no real numbers a and b, not both zero, with au + bv = 0. Because, if one of a, b is not zero in this equation, we can divide by it and hence express one of u, v as a multiple of the other. The concept of direction has an obvious generalization: w has direction u from v (or relative to v) if w−v is a multiple of u. We also say that “w−v has direction u,” and there is no harm in viewing w − v as an abbreviation for the line segment from v to w. As in coordinate geometry, we say that line segments from v to w and from s to t are parallel if they have the same direction; that is, if w − v = a(t − s) for some real number a = 0. Figure 4.3 shows an example of parallel line segments, from v to w and from s to t, both of which have direction u. w = v + 32 u t = s + 12 u
u
s v 0 Figure 4.3: Parallel line segments with direction u Here we have 1 3 w − v = u and t − s = u, 2 2
so w − v = 3(t − s).
70
4 Vectors and Euclidean spaces
Now let us try out the vector concept of parallels on two important theorems from previous chapters. The first is a version of the Thales theorem that parallels cut a pair of lines in proportional segments. Vector Thales theorem. If s and v are on one line through 0, t and w are on another, and w − v is parallel to t − s, then v = as and w = at for some number a. If w − v is parallel to t − s, then w − v = a(t − s) = at − as for some real number a. Because v is on the same line through 0 as s, we have v = bs for some b, and similarly w = ct for some c (this is a good moment to draw a picture). It follows that w − v = ct − bs = at − as, and therefore, (c − a)t + (a − b)s = 0. But s and t are in different directions from 0, hence linearly independent, so c − a = a − b = 0. Thus, v = as and w = at, as required.
As in axiomatic geometry (Exercise 1.4.3), the Pappus theorem follows from the Thales theorem. However, “proportionality” is easier to handle with vectors. Vector Pappus theorem. If r, s, t, u, v, w lie alternately on two lines through 0, with u − v parallel to s − r and t − s parallel to v − w, then u − t is parallel to w − r. Figure 4.4 shows the situation described in the theorem. Because u − v is parallel to s − r, we have u = as and v = ar for some number a. Because t − s is parallel to v − w, we have s = bw and t = bv for some number b. From these two facts, we conclude that u = as = abw and t = bv = bar, hence, u − t = abw − bar = ab(w − r),
4.3 Midpoints and centroids
71
t v=
bar = v =b
ar
r
0
w
s = bw
u = as = abw
Figure 4.4: The parallel Pappus configuration, labeled by vectors and therefore, u − t is parallel to w − r.
The last step in this proof, where we exchange ba for ab, is of course a trifle, because ab = ba for any real numbers a and b. But it is a big step in Chapter 6, where we try to develop geometry without numbers. There we have to build an arithmetic of line segments, and the Pappus theorem is crucial in getting multiplication to behave properly.
Exercises In Chapter 1, we mentioned that a second theorem about parallels, the Desargues theorem, often appears alongside the Pappus theorem in the foundations of geometry. This situation certainly holds in vector geometry, where the appropriate Desargues theorem likewise follows from the vector Thales theorem. 4.2.1 Following the setup explained in Exercise 1.4.4, and the formulation of the vector Pappus theorem above, formulate a “vector Desargues theorem.” 4.2.2 Prove your vector Desargues theorem with the help of the vector Thales theorem.
4.3 Midpoints and centroids The definition of a real vector space does not include a definition of distance, but we can speak of the midpoint of the line segment from u to v and, more generally, of the point that divides this segment in a given ratio.
72
4 Vectors and Euclidean spaces
To see why, first observe that v is obtained from u by adding v − u, the vector that represents the position of v relative to u. More generally, adding any scalar multiple a(v − u) to u produces a point whose direction relative to u is the same as that of v. Thus, the points u + a(v − u) are precisely those on the line through u and v. In particular, the midpoint of the segment between u and v is obtained by adding 12 (v − u) to u, and hence, 1 1 midpoint of line segment between u and v = u + (v − u) = (u + v). 2 2 One might describe this result by saying that the midpoint of the line segment between u and v is the vector average of u and v. This description of the midpoint gives a very short proof of the theorem from Exercise 2.2.1, that the diagonals of a parallelogram bisect each other. By choosing one of the vertices of the parallelogram at 0, we can assume that the other vertices are at u, v, and u + v (Figure 4.5). u+v v u 0 Figure 4.5: Diagonals of a parallelogram Then the midpoint of the diagonal from 0 to u + v is 12 (u + v). And, by the result just proved, this is also the midpoint of the other diagonal—the line segment between u and v. The vector average of two or more points is physically significant because it is the barycenter or center of mass of the system obtained by placing equal masses at the given points. The geometric name for this vector average point is the centroid. In the case of a triangle, the centroid has an alternative geometric description, given by the following classical theorem about medians: the lines from the vertices of a triangle to the midpoints of the respective opposite sides.
4.3 Midpoints and centroids
73
Concurrence of medians. The medians of any triangle pass through the same point, the centroid of the triangle. To prove this theorem, suppose that the vertices of the triangle are u, v, and w. Then the median from u goes to the midpoint 21 (v + w), and so on, as shown in Figure 4.6. w 1 2 (u + w)
1 2 (v + w)
u 1 2 (u + v)
v Figure 4.6: The medians of a triangle Looking at this figure, it seems likely that the medians meet at the point 2/3 of the way from u to 12 (v + w), that is, at the point 2 u+ 3
1 1 2 1 (v + w) − u = u + (v + w) − u = (u + v + w). 2 3 3 3
Voil`a! This is the centroid, and a similar argument shows that it lies 2/3 of the way between v and 12 (u + w) and 2/3 of the way between w and 1 2 (u + v). That is, the centroid is the common point of all three medians. You can of course check by calculation that 13 (u + v + w) lies 2/3 of the way between v and 12 (u + w) and also 2/3 of the way between w and 1 2 (u + v). But the smart thing is not to do the calculation but to predict the result. We know that calculating the point 2/3 of the way between u and 1 2 (v + w) gives 1 (u + v + w), 3 a result that is unchanged when we permute the letters u, v, and w. The other two calculations are the same, except for the ordering of the letters u, v, and w. Hence, they lead to the same result.
74
4 Vectors and Euclidean spaces
Exercises 4.3.1 Show that a square with vertices t, u, v, w has center 14 (t + u + v + w). The theorem about concurrence of medians generalizes beautifully to three dimensions, where the figure corresponding to a triangle is a tetrahedron: a solid with four vertices joined by six lines that bound the tetrahedron’s four triangular faces (Figure 4.7).
Figure 4.7: A tetrahedron 4.3.2 Suppose that the tetrahedron has vertices t, u, v, and w. Show that the centroid of the face opposite to t is 13 (u + v + w), and write down the centroids of the other three faces. 4.3.3 Now consider each line joining a vertex to the centroid of the opposite face. In particular, show that the point 3/4 of the way from t to the centroid of the opposite face is 14 (t + u + v + w)—the centroid of the tetrahedron. 4.3.4 Explain why the point 14 (t + u + v + w) lies on the other three lines from a vertex to the centroid of the opposite face. 4.3.5 Deduce that the four lines from vertex to centroid of opposite face meet at the centroid of the tetrahedron.
4.4 The inner product If u = (u1 , u2 ) and v = (v1 , v2 ) are vectors in R2 , we define their inner product u · v to be u1 v1 + u2 v2 . Thus, the inner product of two vectors is not another vector, but a real number or “scalar.” For this reason, u · v is also called the scalar product of u and v.
4.4 The inner product
75
It is easy to check, from the definition, that the inner product has the algebraic properties u · v = v · u, u · (v + w) = u · v + u · w, (au) · v = u · (av) = a(u · v), which immediately give information about length and angle: • The length |u| is the distance of u = (u1 , u2 ) from 0, which is by the definition of distance in R2 (Section 3.3). Hence,
u21 + u22
|u|2 = u21 + u22 = u · u. It follows that the square of the distance |v − u| from u to v is |v − u|2 = (v − u) · (v − u) = |u|2 + |v|2 − 2u · v. • Vectors u and v are perpendicular if and only if u · v = 0. Because u has slope u2 /u1 and v has slope v2 /v1 , and we know from Section 3.5 that they are perpendicular if and only the product of their slopes is −1. That means v1 u2 =− and hence u2 v2 = −u1 v1 , u1 v2 multiplying both sides by u1 v2 . This equation holds if and only if 0 = u1 v1 + u2 v2 = u · v. We will see in the next section how to extract more information about angle from the inner product. The formula above for |v−u| 2 turns out to be the “cosine rule” or “law of cosines” from high-school trigonometry. But even the criterion for perpendicularity gives a simple proof of a far-fromobvious theorem: Concurrence of altitudes. In any triangle, the perpendiculars from the vertices to opposite sides (the altitudes) have a common point. To prove this theorem, take 0 at the intersection of two altitudes, say those through the vertices u and v (Figure 4.8). Then it remains to show that the line from 0 to the third vertex w is perpendicular to the side v − u.
76
4 Vectors and Euclidean spaces v
0
u
w Figure 4.8: Altitudes of a triangle
Because u is perpendicular to the opposite side w − v, we have u · (w − v) = 0,
that is, u · w − u · v = 0.
Because v is perpendicular to the opposite side u − w, we have v · (u − w) = 0,
that is, v · u − v · w = 0.
Adding these two equations, and bearing in mind that u · v = v · u, we get u · w − v · w = 0,
that is, w · (v − u) = 0.
Thus, w is perpendicular to v − u, as required.
Exercises The inner product criterion for directions to be perpendicular, namely that their inner product is zero, gives a neat way to prove the theorem in Exercise 2.2.2 about the diagonals of a rhombus. 4.4.1 Suppose that a parallelogram has vertices at 0, u, v, and u + v. Show that its diagonals have directions u + v and u − v. 4.4.2 Deduce from Exercise 4.4.1 that the inner product of these directions is |u|2 − |v|2 , and explain why this is zero for a rhombus. The inner product also gives a concise way to show that the equidistant line of two points is the perpendicular bisector of the line connecting them (thus proving more than we did in Section 3.3).
4.5 Inner product and cosine
77
4.4.3 The condition for w to be equidistant from u and v is (w − u) · (w − u) = (w − v) · (w − v). Explain why, and show that this condition is equivalent to |u|2 − 2w · u = |v|2 − 2w · v. 4.4.4 Show that the condition found in Exercise 4.4.3 is equivalent to u+v w− · (u − v) = 0, 2 and explain why this says that w is on the perpendicular bisector of the line from u to v. Having established that the line equidistant from u and v is the perpendicular bisector, we conclude that the perpendicular bisectors of the sides of a triangle are concurrent—because this is obviously true of the equidistant lines of its vertices.
4.5 Inner product and cosine The inner product of vectors u and v depends not only on their lengths |u| and |v| but also on the angle θ between them. The simplest way to express its dependence on angle is with the help of the cosine function. We write the cosine as a function of angle θ , cos θ . But, as usual, we avoid measuring angles and instead define cos θ as the ratio of sides of a rightangled triangle. For simplicity, we assume that the triangle has vertices 0, u, and v as shown in Figure 4.9. v |v|
0
θ
u
|u|
Figure 4.9: Cosine as a ratio of lengths Then the side v is the hypotenuse, θ is the angle between the side u and the hypotenuse, and its cosine is defined by cos θ =
|u| . |v|
78
4 Vectors and Euclidean spaces
We can now use the inner product criterion for perpendicularity to derive the following formula for inner product. Inner product formula. If θ is the angle between vectors u and v, then u · v = |u||v| cos θ . This formula follows because the side v − u of the triangle is perpendicular to side u; hence, 0 = u · (v − u) = u · v − u · u. Therefore, u · v = u · u = |u|2 = |u||v| |u| |v| = |u||v| cos θ .
This formula gives a convenient way to calculate the angle (or at least its cosine) between any two lines, because we know from Section 4.4 how to calculate |u| and |v|. It also gives us the “cosine rule” of trigonometry directly from the calculation of (u − v) · (u − v). Cosine rule. In any triangle, with sides u, v, and u − v, and angle θ opposite to the side u − v, |u − v|2 = |u|2 + |v|2 − 2|u||v| cos θ . Figure 4.10 shows the triangle and the relevant sides and angle, but the proof is a purely algebraic consequence of the inner product formula. v
| −v
0
|u
|v|
θ |u|
u
Figure 4.10: Quantities mentioned in the cosine rule The algebra is simply the following: |u − v|2 = (u − v) · (u − v) = u · u − 2u · v + v · v = |u|2 + |v|2 − 2u · v = |u|2 + |v|2 − 2|u||v| cos θ .
4.5 Inner product and cosine
79
A nice way to close this circle of ideas is to consider the special case in which u and v are the sides of a right-angled triangle and u − v is the hypotenuse. In this case, u is perpendicular to v, so u · v = 0, and the cosine rule becomes hypotenuse2 = |u − v|2 = |u|2 + |v|2 —which is the Pythagorean theorem. This result should not be a surprise, however, because we have already seen how the Pythagorean theorem is built into the definition of distance in R2 and hence into the inner product.
Exercises The Pythagorean theorem can also be proved directly, by choosing 0 at the right angle of a right-angled triangle whose other two vertices are u and v. 4.5.1 Show that |v − u|2 = |u|2 + |v|2 under these conditions, and explain why this is the Pythagorean theorem. While on the subject of right-angled triangles, we mention a useful formula for studying them. 4.5.2 Show that (v + u) · (v − u) = |v|2 − |u|2 . This formula gives a neat proof of the theorem from Section 2.7 about the angle in a semicircle. Take a circle with center 0 and a diameter with ends u and −u as shown in Figure 4.11. Also, let v be any other point on the circle.
v
−u
u 0 Figure 4.11: Points on a semicircle
4.5.3 Show that the sides of the triangle meeting at v have directions v + u and v − u and hence show that they are perpendicular.
80
4 Vectors and Euclidean spaces
4.6 The triangle inequality In vector geometry, the triangle inequality |u + v| ≤ |u| + |v| of Exercises 3.3.1 to 3.3.3 is usually derived from the fact that |u · v| ≤ |u||v|. This result, known as the Cauchy–Schwarz inequality, follows easily from the formula in the previous section. The inner product formula says u · v = |u||v| cos θ , and therefore, |u · v| ≤ |u||v|| cos θ | ≤ |u||v| because | cos θ | ≤ 1. Now, to get the triangle inequality, it suffices to show that |u + v| 2 ≤ (|u| + |v|)2 , which we do as follows: |u + v|2 = (u + v) · (u + v) = |u|2 + 2u · v + |v|2 ≤ |u|2 + 2|u||v| + |v|2 = (|u| + |v|)
because u · u = |u|2 and v · v = |v|2 by Cauchy–Schwarz
2
The reason for the fuss about the Cauchy–Schwarz inequality is that it holds in spaces more complicated than R2 , with more complicated inner products. Because the triangle inequality follows from Cauchy–Schwarz, it too holds in these complicated spaces. We are mainly concerned with the geometry of the plane, so we do not need complicated spaces. However, it is worth saying a few words about Rn , because linear algebra works just as well there as it does in R2 .
Higher dimensional Euclidean spaces Rn is the set of ordered n-tuples (x1 , x2 , . . . , xn ) of real numbers x1 , x2 , . . . , xn . These ordered n-tuples are called n-dimensional vectors. If u and v are in Rn , then we define the vector sum u + v by u + v = (u1 + v1 , u2 + v2 , . . . , un + vn ),
4.6 The triangle inequality
81
and the scalar multiple au for a real number a by au = (au1 , au2 , . . . , aun ). It is easy to check that Rn has the properties enumerated at the beginning of Section 4.1. Hence, Rn is a real vector space under the vector sum and scalar multiplication operations just described. Rn becomes a Euclidean space when we give it the extra structure of an inner product with the properties enumerated in Section 4.4. These properties hold if we define the inner product u · v by u · v = u1 v1 + u2 v2 + · · · + un vn , as is easy to check. This inner product enables us to define distance in R n by the formula |u|2 = u · u which gives the distance |u| of u from the origin. This result is compatible with the concept of distance in R2 or R3 given by the Pythagorean theorem. For example, the distance of (u1 , u2 , u3 ) from 0 in R3 is |u| = u21 + u22 + u23 , as Figure 4.12 shows.
2 + u2
0
u = (u1 , u2 , u3 )
2
u3
u2 1+ u12 + u 2
u3
2
u1
Figure 4.12: Distance in R3
u2
82
4 Vectors and Euclidean spaces •
u21 + u22 is the distance from 0 of (u1 , u2 , 0) (the hypotenuse of a right-angled triangle with sides u1 and u2 ),
• u21 + u22 + u23 is the distance from 0 of (u1 , u2 , u3 ) (the hypotenuse of a right-angled triangle with sides u21 + u22 and u3 ). All theorems proved in this chapter for vectors in the plane R 2 hold in This fact is clear if we take the plane in Rn to consist of vectors of the form (x1 , x2 , 0, . . . , 0), because such vectors behave exactly the same as vectors (x1 , x2 ) in R2 . But in fact any given plane in Rn behaves the same as the special plane of vectors (x1 , x2 , 0, . . . , 0). We skip the details, but it can be proved by constructing an isometry of Rn mapping the given plane onto the special plane. As in R2 , any isometry is a product of reflections. In Rn , at most n + 1 reflections are required, and the proof is similar to the one given in Section 3.7. Rn .
Exercises A proof of Cauchy–Schwarz using only general properties of the inner product can be obtained by an algebraic trick with quadratic equations. The general properties involved are the four listed at the beginning of Section 4.4 and the assumption that w · w = |w|2 ≥ 0 for any vector w (an inner product with the latter property is called positive definite). 4.6.1 The Euclidean inner product for Rn defined above is positive definite. Why? 4.6.2 For any real number x, and any vectors u and v, show that (u + xv) · (u + xv) = |u|2 + 2x(u · v) + x2 |v|2 , and hence that |u|2 + 2x(u · v) + x2 |v|2 ≥ 0 for any real number x. 4.6.3 If A, B, and C are real numbers and A + Bx +Cx2 ≥ 0 for any real number x, explain why B2 − 4AC ≤ 0. 4.6.4 By applying Exercise 4.6.3 to the inequality |u|2 + 2x(u · v) + x2 |v|2 ≥ 0, show that (u · v)2 ≤ |u|2 |v|2 ,
and hence
|u · v| ≤ |u||v|.
4.7 Rotations, matrices, and complex numbers
83
4.7 Rotations, matrices, and complex numbers Rotation matrices In Section 3.6, we defined a rotation of R2 as a function rc,s , where c and s are two real numbers such that c2 +s2 = 1. We described rc,s as the function that sends (x, y) to (cx − sy, sx + cy), but it is also described by the matrix of coefficients of x and y, namely
c −s s c
,
where c = cos θ and s = sin θ .
Because most readers will already have seen matrices, it may be useful to translate some previous statements about functions into matrix language, where they may be more familiar. (Readers not yet familiar with matrices will find an introduction in Section 7.2.) Matrix notation allows us to rewrite (x, y) → (cx − sy, sx + cy) as
c −s s c
x y
=
cx − sy sx + cy
Thus, the function rc,s is applied to the variablesx and y bymultiplying the x c −s column vector on the left by the matrix . Functions are y s c thereby separated from their variables, so they can be composed without the variables becoming involved—simply by multiplying matrices. This idea gives proofs of the formulas for cos(θ1 + θ2 ) and sin(θ1 + θ2 ), similar to Exercises 3.5.3 and 3.5.4, but with the variables x and y filtered out: • Rotation through angle θ1 is given by the matrix • Rotation through angle θ2 is given by the matrix
cos θ1 − sin θ1 sin θ1 cos θ1 cos θ2 − sin θ2 sin θ2 cos θ2
. .
• Hence, rotation through θ1 + θ2 is given by the product of these two matrices. That is,
84
4 Vectors and Euclidean spaces
cos(θ1 + θ2 ) − sin(θ1 + θ2 ) sin(θ1 + θ2 ) cos(θ1 + θ2 ) cos θ2 − sin θ2 cos θ1 − sin θ1 = sin θ1 cos θ1 sin θ2 cos θ2 cos θ1 cos θ2 − sin θ1 sin θ2 − cos θ1 sin θ2 − sin θ1 cos θ2 = cos θ1 sin θ2 + sin θ1 cos θ2 cos θ1 cos θ2 − sin θ1 sin θ2
by matrix multiplication. • Finally, equating corresponding entries in the first and last matrices, cos(θ1 + θ2 ) = cos θ1 cos θ2 − sin θ1 sin θ2 , sin(θ1 + θ2 ) = cos θ1 sin θ2 + sin θ1 cos θ2 .
Complex numbers One advantage of matrices, which we do not pursue here, is that they can be used to generalize the idea of rotation to any number of dimensions. But, for rotations of R2 , there is a notation even more efficient than the rotation matrix cos θ − sin θ . sin θ cos θ √ It is the complex number cos θ + i sin θ , where i = −1. We represent the point (x, y) ∈ R2 by the complex number z = x + iy, and we rotate it through angle θ about O by multiplying it by cos θ +i sin θ . This procedure works because i2 = −1, and therefore, (cos θ + i sin θ )(x + iy) = x cos θ − y sin θ + i(x sin θ + y cos θ ). Thus, multiplication by cos θ + i sin θ sends each point (x, y) to the point (x cos θ −y sin θ , x sin θ +y cos θ ), which is the result of rotating (x, y) about O through angle θ . Multiplying all points at once by cos θ + i sin θ , therefore, rotates the whole plane about O through angle θ . It follows that multiplication by (cos θ1 + i sin θ1 )(cos θ2 + i sin θ2 ) rotates the plane through θ1 + θ2 —the first factor rotates it through θ1 and the second rotates it through θ2 —so it is the same as multiplication by
4.7 Rotations, matrices, and complex numbers
85
cos(θ1 + θ2 ) + i sin(θ1 + θ2 ). Equating these two multipliers gives perhaps the ultimate proof of the formulas for cos(θ1 + θ2 ) and sin(θ1 + θ2 ): cos(θ1 + θ2 )+i sin(θ1 + θ2 ) = (cos θ1 + i sin θ1 )(cos θ2 + i sin θ2 ) = cos θ1 cos θ2 − sin θ1 sin θ2 + i(cos θ1 sin θ2 + sin θ1 cos θ2 ) since i2 = −1. Hence, equating real and imaginary parts, cos(θ1 + θ2 ) = cos θ1 cos θ2 − sin θ1 sin θ2 , sin(θ1 + θ2 ) = cos θ1 sin θ2 + sin θ1 cos θ2 .
Exercises The calculations above show that multiplication by cos θ + i sin θ is rotation about O through angle θ because of the (seemingly accidental) property i2 = −1. In fact, any algebra of points in R2 that satisfies the same laws as the algebra of R automatically satisfies the condition i2 = −1, where i is the point (0, 1). The following exercises show why. In particular, they reveal geometric consequences of the following algebraic laws: |uv| = |u||v|
(multiplicative absolute value)
u(v + w) = uv + uw (distributive law) 4.7.1 Given that |x + iy| = x2 + y2 , explain why |v − w| equals the distance between the complex numbers v and w. 4.7.2 Assuming the multiplicative absolute value and the distributive law (and, if necessary, any other algebraic laws satisfied by R), show that distance between uv and uw = |u| × distance between v and w. In other words, multiplying the plane C of complex numbers by a constant complex number u multiplies all distances by |u|. 4.7.3 Deduce from Exercises 4.7.1 and 4.7.2 that multiplication of C by a number u with |u| = 1 is an isometry leaving O fixed. 4.7.4 Assuming that u = 1, and hence that uz = z when z = 0, deduce from Exercise 4.7.3 that multiplication by u = 1 is a rotation. These results explain why multiplication by u with |u| = 1 is a rotation. To find the angle of rotation we assume that the point (1, 0) is the 1 of the algebra and observe where the rotation sends 1.
86
4 Vectors and Euclidean spaces
4.7.5 Explain why any u with |u| = 1 can be written in the form cos θ + i sin θ for some angle θ , and conclude that multiplication by u rotates the point 1 (hence the whole plane) through angle θ . It follows, in particular, that multiplication by i = (0, 1) sends (1, 0) to (0, 1) and hence rotates the plane through π /2. This result in turn implies i2 = −1, because multiplication by i2 then rotates the plane through π , which is also the effect of multiplication by −1.
4.8 Discussion Because the geometric content of a vector space with an inner product is much the same as Euclidean geometry, it is interesting to see how many axioms it takes to describe a vector space. Remember from Section 2.9 that it takes 17 Hilbert axioms to describe the Euclidean plane, or 16 if we are willing to drop completeness of the line. To define a vector space, we began in Section 4.1 with eight axioms for vector addition and scalar multiplication: u+v = v+u
1u = u
u + (v + w) = (u + v) + w
a(u + v) = au + av
u+0 = u
(a + b)u = au + bu
u + (−u) = 0
a(bu) = (ab)u.
Then, in Section 4.4, we added three (or four, depending on how you count) axioms for the inner product: u · v = v · u, u · (v + w) = u · v + u · w, (au) · v = u · (av) = a(u · v), We also need relations among inner product, length, and angle—at a minimum the cosine formula, u · v = |u||v| cos θ , so this is 12 or 13 axioms so far. But we have also assumed that the scalars a, b, . . . are real numbers, so there remains the problem of writing down axioms for them. At the very
4.8 Discussion
87
least, one needs axioms saying that the scalars satisfy the ordinary rules of calculation, the so-called field axioms (this is usual when defining a vector space): a + b = b + a,
ab = ba
a + (b + c) = (a + b) + c,
a(bc) = (ab)c
a + 0 = a, a + (−a) = 0,
a1 = a aa a(b + c) = ab + ac
−1
=1
(commutative laws) (associative laws) (identity laws) (inverse laws) (distributive law)
Thus, the usual definition of a vector space, with an inner product suitable for Euclidean geometry, takes more than 20 axioms! Admittedly, the field axioms and the vector space axioms are useful in many other parts of mathematics, whereas most of the Hilbert axioms seem meaningful only in geometry. And, by varying the inner product slightly, one can change the geometry of the vector space in interesting ways. For example, one can obtain the geometry of Minkowski space used in Einstein’s special theory of relativity. To learn more about the vector space approach to geometry, see Linear Algebra and Geometry, a Second Course by I. Kaplansky and Metric Affine Geometry by E. Snapper and R. J. Troyer. Still, one can dream of building geometry on a much simpler set of axioms. In Chapter 6, we will realize this dream with projective geometry, which we begin studying in Chapter 5.
5 Perspective P REVIEW Euclid’s geometry concerns figures that can be drawn with straightedge and compass, even though many of its theorems are about straight lines alone. Are there any interesting figures that can be drawn with straightedge alone? Remember, the straightedge has no marks on it, so it is impossible to copy a length. Thus, with a straightedge alone, we cannot draw a square, an equilateral triangle, or any figure involving equal line segments. Yet there is something interesting we can draw: a perspective view of a tiled floor, such as the one shown in Figure 5.1.
Figure 5.1: Perspective view of a tiled floor This picture is interesting because it seems clear that all tiles in the view are of equal size. Thus, even though we cannot draw tiles that are actually equal, we can draw tiles that look equal. We will explain how to solve the problem of drawing perspective views in Section 5.2. The solution takes us into a new form of geometry—a geometry of vision—called projective geometry.
88
5.1 Perspective drawing
89
5.1 Perspective drawing Sometime in the 15th century, Italian artists discovered how to draw threedimensional scenes in correct perspective. Figures 5.2 and 5.3 illustrate the great advance in realism this skill achieved, with pictures drawn before and after the discovery. The “before” picture, Figure 5.2, is a drawing I found in the book Perspective in Perspective by L. Wright. It is thought to date from the late 15th century, but it comes from England, where knowledge of perspective had evidently not reached at that time.
Figure 5.2: The birth of St Edmund, by an unknown artist The “after” picture, Figure 5.3, is the 1514 engraving Saint Jerome in his study, by the great German artist Albrecht Du¨ rer (1471–1528). D¨urer made study tours of Italy in 1494 and 1505 and became a master of all aspects of drawing, including perspective. The simplest test of perspective drawing is the depiction of a tiled floor. The picture in Figure 5.2 clearly fails this test. All the tiles are drawn as rectangles, which makes the floor look vertical. We know from experience that a horizontal rectangle does not look rectangular—its angles are not all right angles because its sides converge to a common point on the horizon, as in the tabletop in D¨urer’s engraving.
90
5 Perspective
Figure 5.3: St Jerome in his study, by Albrecht Du¨ rer
The Italians drew tiles by a method called the costruzione legittima (legitimate construction), first published by Leon Battista Alberti in 1436. The bottom edge of the picture coincides with a line of tile edges, and any other horizontal line is chosen as the horizon. Then lines drawn from equally spaced points on the bottom edge to a point on the horizon depict the parallel columns of tiles perpendicular to the bottom edge (Figure 5.4). Another horizontal line, near the bottom, completes the first row of tiles.
Figure 5.4: Beginning the costruzione legittima
5.1 Perspective drawing
91
The real problem comes next. How do we find the correct horizontal lines to depict the 2nd, 3rd, 4th, . . . rows of tiles? The answer is surprisingly simple: Draw the diagonal of any tile in the bottom row (shown in gray in Figure 5.5). The diagonal necessarily crosses successive columns at the corners of tiles in the 2nd, 3rd, 4th, . . . rows; hence, these rows can be constructed by drawing horizontal lines at the successive crossings. Voil`a!
Figure 5.5: Completing the costruzione legittima
Exercises Suppose that the floor has rows of tiles crossing the x-axis at x = 0, 1, 2, 3, . . ., and that the artist copies the view of the floor onto a vertical transparent screen through the y-axis, keeping a fixed eye position at the point (−1, 1). Then the perspective view of the points x = 0, 1, 2, 3, . . . will be the series of points on the y-axis shown in Figure 5.6.
y
O
1
2
3
4
x
Figure 5.6: Perspective view of equally spaced points n 5.1.1 Show that the line from (−1, 1) to (n, 0) crosses the y-axis at y = n+1 . Hence, the perspective images of the points x = 0, 1, 2, 3, . . . are the points y = 0, 12 , 23 , 34 , . . ..
92
5 Perspective
If each of the points 0, 1, 2, 3, . . . is sent to the next, then each of their perspective images y = 0, 12 , 23 , 34 , . . . is sent to the next. 5.1.2 Show that the function f (y) =
1 2−y
effects this move.
5.1.3 Which point on the y-axis is not moved by the function f (y) = what is the geometric significance of this point?
1 2−y ,
and
5.2 Drawing with straightedge alone The construzione legittima takes advantage of something that is visually obvious but mathematically mysterious—the fact that parallel lines generally do not look parallel, but appear to meet on the horizon. The point where a family of parallels appear to meet is called their “vanishing point” by artists, and their point at infinity by mathematicians. The horizon itself, which consists of all the points at infinity, is called the line at infinity. However, the costruzione legittima does not take full advantage of points at infinity. It involves some parallels that are really drawn parallel, so we need both straightedge and compass as used in Chapter 1. The construction also needs measurement to lay out the equally spaced points on the bottom line of the picture, and this again requires a compass. Thus, the costruzione legittima is a Euclidean construction at heart, requiring both a straightedge and a compass. Is it possible to draw a perspective view of a tiled floor with a straightedge alone? Absolutely! All one needs to get started is the horizon and a tile placed obliquely. The tile is created by the two pairs of parallel lines, which are simply pairs that meet on the horizon (Figure 5.7). horizon
Figure 5.7: The first tile We then draw the diagonal of this tile and extend it to the horizon, obtaining the point at infinity of all diagonals parallel to this first one. This step allows us to draw two more diagonals, of tiles adjacent to the first one.
5.2 Drawing with straightedge alone
93
These diagonals give us the remaining sides of the adjacent tiles, and we can then repeat the process. The first few steps are shown in Figure 5.8. Figure 5.1 at the beginning of the chapter is the result of carrying out many steps (and deleting the construction lines).
Draw diagonal of first tile, extended to the horizon
Extend diagonal of second tile to the horizon
Draw side of second tile, through the new intersection
Draw side of more tiles, through the new intersection
Figure 5.8: Constructing the tiled floor
This construction is easy and fun to do, and we urge the reader to get a straightedge and try it. Also try the constructions suggested in the exercises, which create pictures of floors with differently shaped tiles.
94
5 Perspective
Exercises Consider the triangular tile shown shaded in Figure 5.9. Notice that this triangle could be half of the quadrangular tile shown in Figure 5.7 (this is a hint).
Figure 5.9: A triangular tile 5.2.1 Draw a perspective view of the plane filled with many copies of this tile. 5.2.2 Also, by deleting some lines in your solution to Exercise 5.2.1, create a perspective view of the plane filled with congruent hexagons.
5.3 Projective plane axioms and their models Drawing a tiled floor with straightedge alone requires a “horizon”—a line at infinity. Apart from this requirement, the construction works because certain things remain the same in any view of the plane: • straight lines remain straight • intersections remain intersections • parallel lines remain parallel or meet on the horizon. Now parallel lines always meet on the horizon if you point yourself in the right direction, so if we could look in all directions at once we would see that any two lines have a point in common. This idea leads us to believe in a structure called a projective plane, containing objects called “points” and “lines” satisfying the following axioms. We write “points” and “lines” in quotes because they may not be the same as ordinary points and lines. Axioms for a projective plane 1. Any two “points” are contained in a unique “line.” 2. Any two “lines” contain a unique “point.” 3. There exist four “points”, no three of which are in a “line.”
5.3 Projective plane axioms and their models
95
Notice that these are axioms about incidence: They involve only meetings between “points” and “lines,” not things such as length or angle. Some of Euclid’s and Hilbert’s axioms are of this kind, but not many. Axiom 1 is essentially Euclid’s first axiom for the construction of lines. Axiom 2 says that there are no exceptional pairs of lines that do not meet. We can define “parallels” to be lines that meet on a line called the “horizon,” but this does not single out a special class of lines—in a projective plane, the “horizon” behaves the same as any other line. Axiom 3 says that a projective plane has “enough points to be interesting.” We can think of the four points as the four vertices of a quadrilateral, from which one may generate the complicated structure seen in the pictures of a tiled floor at the beginning of this chapter. The real projective plane If there is such a thing as a projective plane, it should certainly satisfy these axioms. But does anything satisfy them? After all, we humans can never see all of the horizon at once, so perhaps it is inconsistent to suppose that all parallels meet. These doubts are dispelled by the following model, or interpretation, of the axioms for a projective plane. The model is called the real projective plane RP2 , and it gives a mathematical meaning to the terms “point,” “line,” and “plane” that makes all the axioms true. Take “points” to be lines through O in R3 , “lines” to be planes through O in R3 , and the “plane” to be the set of all lines through O in R3 . Then 1. Any two “points” are contained in a unique “line” because two given lines through O lie in a unique plane through O. 2. Any two “lines” contain a unique “point” because any two planes through O meet in a unique line through O. 3. There are four different “points,” no three of which are in a “line”: for example, the lines from O to the four points (1, 0, 0), (0, 1, 0), (0, 0, 1), and (1, 1, 1), because no three of these lines lie in the same plane through O. The last claim is perhaps a little hard to grasp by visualization, but it can be checked algebraically because any plane through O has an equation of the form ax + by + cz = 0 for some real numbers a, b, c.
96
5 Perspective
If, say, (1, 0, 0) and (0, 1, 0) are on this plane, then we find by substituting these values of x, y, z in the equation that a = 0 and b = 0,
hence the plane is z = 0.
But (0, 0, 1) and (1, 1, 1) do not lie on the plane z = 0. It can be checked similarly that the plane through any two of the points does not contain the other two. It is no fluke that lines and planes through O in R3 behave as we want “points” and “lines” of a projective plane to behave, because they capture the idea of viewing with an all-seeing eye. The point O is the position of the eye, and the lines through O connect the eye to points in the plane. Consider how the eye sees the plane z = −1, for example (Figure 5.10). z
y x L1 M
L3
L2
P1
P2
P3
Figure 5.10: Viewing a plane from O Points P1 , P2 , P3 , . . . in the plane z = −1 are joined to the eye by lines L1 , L2 , L3 , . . . through O, and as the point Pn tends to infinity, the line Ln tends toward the horizontal. Therefore, it is natural to call the horizontal lines through O the “points at infinity” of the plane z = −1, and to call the plane of all horizontal lines through O the “horizon” or “line at infinity” of the plane z = −1. Unlike the lines L1 , L2 , L3 , . . ., corresponding to points P1 , P2 , P3 , . . . of the Euclidean plane z = −1, horizontal lines through O have no counterparts in the Euclidean plane: They extend the Euclidean plane to a projective plane. However, the extension arises in a natural way. Once we
5.3 Projective plane axioms and their models
97
replace the points P1 , P2 , P3 , . . . by lines in space, we realize that there are extra lines (the horizontal lines) corresponding to the points on the horizon. This model of the projective plane nicely captures our intuitive idea of points at infinity, but it also makes the idea clearer. We can see, for example, why it is proper for each line to have only one point at infinity, not two: because the lines L connecting O to points P along a line M in the plane z = −1 tend toward the same horizontal line as P tends to infinity in either direction (namely, the parallel to M through O). It is hard to find a surface that behaves like RP2 , but it is easy to find a curve that behaves like any “line” in it, a so-called real projective line. Figure 5.11 shows how. The “points” in a “line” of RP 2 , namely the lines through O in some plane through O, correspond to points of a circle through O. Each point P = O on the circle corresponds to the line through O and P, and the point O itself corresponds to the tangent line at O. O P
Figure 5.11: Modeling a projective line by a circle
Exercises To gain more familiarity with calculations in R3 , let us pursue the example of four “points” given above. 5.3.1 Find the plane ax + by + cz = 0 through the points (0, 0, 1) and (1, 1, 1), and check that it does not contain the points (1, 0, 0) and (0, 1, 0). 5.3.2 Show that RP2 has four “lines,” no three of which have a common “point.” Not only does RP2 contain four “lines,” no three of which have a “point” in common; the same is true of any projective plane, because this property follows from the projective plane axioms alone. 5.3.3 Suppose that A, B,C, D are four “points” in a projective plane, no three of which are in a “line.” Consider the “lines” AB, BC,CD, DA. Show that if AB and BC have a common point E, then E = B. 5.3.4 Deduce from Exercise 5.3.3 that the three lines AB, BC,CD have no common point, and that the same is true of any three of the lines AB, BC,CD, DA.
98
5 Perspective
5.4 Homogeneous coordinates Because “points” and “lines” of RP2 are lines and planes through O in R3 , they are easily handled by methods of linear algebra. A line through O is determined by any point (x, y, z) = O, and it consists of the points (tx,ty,tz), where t runs through all real numbers. Thus, a “point” is not given by a single triple (x, y, z), but rather by any of its nonzero multiples (tx,ty,tz). These triples are called the homogeneous coordinates of the “point.” A plane through O has a linear equation of the form ax + by + cz = 0, called a homogeneous equation. The same plane is given by the equation tax+tby+tcz = 0 for any nonzero t. Thus, a “line” is likewise not given by a single triple (a, b, c), but by the set of all its nonzero multiples (ta,tb,tc). If (x1 , y1 , z1 ) and (x2 , y2 , z2 ) lie on different lines through O, then it is geometrically obvious that they lie in a unique plane ax + by + cz = 0. The coordinates (a, b, c) of this plane can be found by solving the two equations ax1 + by1 + cz1 = 0, ax2 + by2 + cz2 = 0, for a, b, and c. Because there are more unknowns than equations, there is not a single solution triple but a whole space of them—in this case, a set of multiples (ta,tb,tc), all representing the same homogeneous equation. This is the algebraic reason why two “points” lie on a unique “line” in 2 RP . There is a similar reason why two “lines” have a unique “point” in common. Two “lines” are given by two equations a1 x + b1 y + c1 z = 0, a2 x + b2 y + c2 z = 0, and we find their common “point” by solving these equations for x, y, and z. This problem is the same as above, but with the roles of a, b, c exchanged with those of x, y, z. The solution in this case is a set of multiples (tx,ty,tz) representing the homogeneous coordinates of the common “point.” The practicalities of finding the “line” through two “points” or the “point” common to two “lines” are explored in the next exercise set. But first I want to make a theoretical point. It makes no algebraic difference if the coordinates of “points” and “lines” are complex numbers. We can define a complex projective plane CP2 , each “point” of which is a set of triples of the form (tx,ty,tz), where x, y, z are particular complex numbers
5.4 Homogeneous coordinates
99
and t runs through all complex numbers. It remains true that any two “points” lie on a unique “line” and any two “lines” have unique common point, simply because the algebraic properties of complex linear equations are exactly the same as those of real linear equations. Similarly, one can show there are four “points,” no three of which are in a “line” of CP 2 . Thus, there is more than one model of the projective plane axioms. Later we shall look at other models, which enable us to see that certain properties of RP2 are not properties of all projective planes and hence do not follow from the projective plane axioms.
Projective space It is easy to generalize homogeneous coordinates to quadruples (w, x, y, z) and hence to define the three-dimensional real projective space RP 3 . It has “points,” “lines,” and “planes” defined as follows (we use vector notation to shorten the definitions): • A “point” is a line through O in R4 , that is, a set of quadruples tu, where u = (w, x, y, z) is a particular quadruple of real numbers and t runs through all real numbers. • A “line” is a plane through O in R4 , that is, a set t1 u1 + t2 u2 where u1 and u2 are linearly independent points of R4 and t1 and t2 run through all real numbers. • A “plane” is a three-dimensional space through O in R4 , that is, a set t1 u1 + t2 u2 + t3 u3 , where u1 , u2 , and u3 are linearly independent points of R4 and t1 , t2 , and t3 run through all real numbers. Linear algebra then enables us to show various properties of the “points,” “lines,” and “planes” in RP3 , such as: 1. Two “points” lie on a unique “line.” 2. Three “points” not on a “line” lie on a unique “plane.” 3. Two “planes” have unique “line” in common. 4. Three “planes” with no common “line” have one common “point.” These properties hold for any three-dimensional projective space, and RP 3 is not the only one. There is also a complex projective space CP 3 , and many others. RP3 has an unexpected influence on the geometry of the sphere, as we will see in Section 7.8.
100
5 Perspective
Exercises 5.4.1 Find the plane ax + by + cz = 0 that contains the points (1, 2, 3) and (1, 1, 1). 5.4.2 Find the line of intersection of the planes x + 2y + 3z = 0 and x + y + z = 0. 5.4.3 You can write down the solution of Exercise 5.4.2 as soon as you have solved Exercise 5.4.1. Why?
5.5 Projection The three-dimensional Euclidean space R3 , in which the lines through O are the “points” of RP2 and the planes through O are the “lines” of RP2 , also contains many other planes. Each plane P not passing through O can be regarded as a perspective view of the projective plane RP 2 , a view that contains all but one “line” of RP2 . Each point P of P corresponds to a line (“of sight”) through O, and hence to a “point” of RP2 . The only lines through O that do not meet P are those parallel to P, and these make up the line at infinity or horizon of P, as we have already seen in the case of the plane z = −1 in Section 5.3. If P1 and P2 are any two planes not passing through O we can project P1 to P2 by sending each point P1 in P1 to the point P2 in P2 lying on the same line through O as P1 (Figure 5.12). The geometry of RP2 is called “projective” because it encapsulates the geometry of a whole family of planes related by projection.
P2
P1 O
P2
P1
Figure 5.12: Projecting one plane to another
5.5 Projection
101
Projections of projective lines Projection of one plane P1 onto another plane P2 produces an image of P1 that is generally distorted in some way. For example, a grid of squares on P1 may be mapped to a perspective view of the grid that looks like Figure 5.1. Nevertheless, straight lines remain straight under projection, so there are limits to the amount of distortion in the image. To better understand the nature and scope of projective distortion, in this subsection we analyze the mappings of the projective line obtainable by projection. An effective way to see the distortion produced by projection of one line L1 onto another line L2 is to mark a series of equally spaced dots on L1 and the corresponding image dots on L2 . You can think of the image dots as “shadows” of the dots on L1 cast by light rays from the point of projection P, except that we have projective lines through P, not rays, so it can seem as though the “shadow” on L2 comes ahead of the dot on L1 . (See Figure 5.15, but bear in mind that a projective line is really circular, so it is always possible to pass through P, to a point on L1 , then to a point on L2 , in that order.) In the simplest cases, where L1 and L2 are parallel, the image dots are also equally spaced. Figure 5.13 shows the case of projection from a point at infinity, where the lines from the dots on L1 to their images on L2 are parallel and hence the dots on L1 are simply translated a constant distance l. If we choose an origin on each line and use the same unit of length on each, then projection from infinity sends each x on L1 to x + l on L2 .
0
0
1
l
1+l
L1
3
2
2+l
3+l
L2
Figure 5.13: Projection from infinity When L1 is projected from a finite point P, then the distance between dots is magnified by a constant factor k = 0. If we take P on a line through
102
5 Perspective
the zero points on L1 and L2 , then the projection sends each x on L1 to kx on L2 (Figure 5.14). Note also that this projection sends x on L2 to x/k on L1 , so the magnification factor can be any k = 0. P
L1
x
0
kx
0
L2
Figure 5.14: Projection from a finite point When L1 and L2 are not parallel the distortion caused by projection is more extreme. Figure 5.15 shows how the spacing of dots changes when L1 is projected onto a perpendicular line L2 from a point O equidistant from both. Figure 5.16 is a closeup of the image line L2 , showing how the image dots “converge” to a point corresponding to the horizontal line through O (which corresponds to the point at infinity on L1 ). L2
L1
O
Figure 5.15: Example of projective distortion of the line
5.5 Projection
103
− 21
−1
− 31
1 1 5 4
1 3
1 2
1
L2
Figure 5.16: Closeup of the image line We take O = (0, 0) as usual, and we suppose that L1 is parallel to the x-axis, that L2 is parallel to the y-axis, and that the dots on L1 are unit distance apart. Then the line from O to the dot at x = n on L1 has slope 1/n and hence it meets the line L2 at y = 1/n. Thus the map from L1 to L2 is the function sending x to y = 1/x. This map exhibits the most extreme kind of distortion induced by projection, with the point at infinity on L1 sent to the point y = 0 on L2 . Any combination of these projections is therefore a combination of functions 1/x, kx, and x + l, which are called generating transformations. The combinations of generating transformations are precisely the functions of the form ax + b f (x) = , where ad − bc = 0, cx + d that we study in the next section.
Exercises Before studying all these functions, it is useful to study the (simpler) subclass obtained by composing functions that send x to x + l or kx (for k = 0). The latter functions obviously include any function of the form f (x) = ax + b with a = 0, which is the result of multiplying by a, and then adding b. 5.5.1 If f1 (x) = a1 x + b1 with a1 = 0 and f2 (x) = a2 x + b2 with a2 = 0, show that f1 ( f2 (x)) = Ax + B,
with A = 0,
and find the constants A and B. 5.5.2 Deduce from Exercise 5.5.1 that the result of composing any number of functions that send x to x + l or kx (for k = 0) is a function of the form f (x) = ax + b with a = 0. We know that such functions represent combinations of certain projections from lines to parallel lines, but do they include any projection from a line to a parallel line? 5.5.3 Show that projection of a line, from any finite point P, onto a parallel line is represented by a function of the form f (x) = ax + b.
104
5 Perspective
5.6 Linear fractional functions The functions sending x to 1/x, kx, and x + l are among the functions called linear fractional, each of which has the form f (x) =
ax + b cx + d
where ad − bc = 0.
The condition ad − bc = 0 ensures that f (x) is not constant. Constancy occurs only if ax + b = ac (cx + d); in which case, ad − bc = 0 because ad c = b. By writing ax + b cx + d
as
ad a ax + ad (cx + d) + 1c (bc − ad) c +b− c = c cx + d cx + d
we find that any linear fractional function with c = 0 may be written in the form a bc − ad f (x) = + . c c(cx + d) Such a function may therefore be composed from functions sending x to 1/x, kx, and x + l—the functions that reciprocate, multiply by k, and add l—for various values of k and l: • first multiply x by c, • then add d, • then multiply again by c, • then reciprocate, • then multiply by bc − ad, • and finally add ac , a bc−ad ax+b c + c(cx+d) = cx+d . When c = 0, the linear ax+b a b cx+d = d x+ d , and this can be composed from
and the result is that x goes to
fractional function is simply x by multiplying by a/d and then adding b/d. Thus, any linear fractional function is composed from the functions that reciprocate, multiply by k, and add l, and hence (by the constructions in the previous section) any linear fractional function on the number line is realized by a sequence of projections of the line.
5.6 Linear fractional functions
105
We now wish to prove the converse: Any sequence of projections of the number line realizes a linear fractional function. From the previous section, we know this is true for projection of a line onto a parallel line, so it suffices to find the function realized by projection of a line onto an intersecting line. We first take the case in which the lines are perpendicular (Figure 5.17). This case generalizes that of Figure 5.15, by allowing projection from an arbitrary point (a, b). y (a, b)
O t
x
f (t)
Figure 5.17: Projecting a line onto a perpendicular line To find where the point t on the x-axis goes on the y-axis, we consider the slope of the line through t and (a, b). Between these points, the rise is b . Between t and the point f (t) on b and the run is a − t, so the slope is a−t the y-axis, the run is t and the rise is − f (t); hence, bt , which is a linear fractional function. t −a For the general case of intersecting lines, we take one line to be the x-axis again, and the other to be the line y = cx. Again we project the point t on the x-axis from (a, b) to the other line, and to find where t goes, we first find the equation of the line through t and (a, b). Equating the slope from t to (a, b) with the slope between an arbitrary point (x, y) on the line and (a, b), we find the equation f (t) =
b−y b = . a−t a−x This line meets the line y = cx where b b − cx = , a−t a−x
106
5 Perspective
and hence where x=
bt , ct − ac + b
which is also a linear fractional function of t.
Thus, any single projection of a line can be represented by a linear fractional function of distance along the line. It is easy to check (Exercise 5.6.2) that the result of composing linear fractional functions is linear fractional. Hence, any finite sequence of projections is represented by a linear fractional function.
Dividing by zero You remember from high-school algebra that division by zero is not a valid operation, because it leads from true equations, such as 3 × 0 = 2 × 0, to false ones, such as 3 = 2. Nevertheless, in carefully controlled situations, it is permissible, and even enlightening, to divide by zero. One such situation is in projective mappings of the projective line. The linear fractional functions f (x) = ax+b cx+d we have used to describe projective mappings of lines are actually defective if the variable x runs only through the set R of real numbers. For example, the function f (x) = 1/x we used to map points of the line L1 onto points of the line L2 as shown in Figure 5.15 does not in fact map all points. It cannot send the point x = 0 anywhere, because 1/0 is undefined; nor can it send any point to y = 0, because 0 = 1/x for any real x. This defect is neatly fixed by extending the function f (x) = 1/x to a new object x = ∞, and declaring that 1/∞ = 0 and 1/0 = ∞. The new object ∞ is none other than the point at infinity of the line L1 , which is supposed to map to the point 0 on L2 . Likewise, if 1/0 = ∞, the point 0 on L1 is sent to the point ∞ on L2 , as it should be. Thus, the function f (x) = 1/x works properly, not on the real line R, but on the real projective line R ∪ {∞}—a line together with a point at infinity. The rules 1/∞ = 0 and 1/0 = ∞ simply reflect this fact. It is much the same with any linear fractional function f (x) = ax+b cx+d . The denominator of the fraction is 0 when x = −d/c, and the correct value of the function in this case is ∞. Conversely, no real value of x gives f (x) the value a/c, but x = ∞ does. For this reason, any function f (x) = ax+b cx+d with ad − bc = 0 maps the real projective line R ∪ {∞} onto itself. The map is also one-to-one, as may seen in the exercises below.
5.6 Linear fractional functions
107
The real projective line RP1 We can now give an algebraic definition of the object we called the “real projective line” in Section 5.3. It is the set R ∪ {∞} together with all the linear fractional functions mapping R ∪ {∞} onto itself. We call this set, with these functions on it, the real projective line RP1 . The set R∪{∞} certainly has the points we require for a projective line; the functions are to give R ∪ {∞} the “elasticity” of a line that undergoes projection. The ordinary line R is not very “elastic” in this sense. Once we have decided which point is 0 and which point is 1, the numerical value of every point on R is uniquely determined. In contrast, the position of a point on RP1 is not determined by the positions of 0 and 1 alone. For example, there is a projection that sends 0 to 0, 1 to 1, but 2 to 3. Nevertheless, there is a constraint on the “elasticity” of RP 1 . If 0 goes to 0, 1 goes to 1, and 2 goes to 3, say, then the destination of every other point x is uniquely determined. In the next two sections, we will see why.
Exercises The formula
ax+b cx+d
bc−ad = ac + c(cx+d) gives an inkling why the condition ad − bc = 0 is
a part of the definition of a linear fractional function: If ad − bc = 0, then ax+b cx+d = c is a constant function, and hence it maps the whole line onto one point. If we want to map the line onto another line, it is therefore necessary to have ad − bc = 0. It is also sufficient, because we can solve the equation y = ax+b cx+d for x in that case.
5.6.1 Solve the equation y = ad − bc = 0.
ax+b cx+d
for x, and note where your solution assumes
x+b1 5.6.2 If f1 (x) = ac11x+d and f2 (x) = 1 Ax+B is of the form Cx+D .
5.6.3 Verify also that
A C
B D
a2 x+b2 c2 x+d2 ,
=
a1 c1
compute f1 ( f2 (x)), and verify that it
b1 d1
a2 c2
b2 d2
.
Thus, linear fractional functions behave like 2 × 2 matrices. Moreover, the condition ad − bc = 0 corresponds to having nonzero determinant, which explains why this is the condition for an inverse function to exist. 5.6.4 It also guarantees that if a1 d1 − b1 c1 = 0 for f1 (x) and a2 d2 − b2 c2 = 0 for f2 (x) in Exercise 5.5.2, then AD − BC = 0 for f 1 ( f2 (x)). Why?
108
5 Perspective
5.7 The cross-ratio You might say it was a triumph of algebra to invent this quantity that turns out to be so valuable and could not be imagined geometrically. Or if you are a geometer at heart, you may say it is an invention of the devil and hate it all your life. Robin Hartshorne, Geometry: Euclid and Beyond, p. 341. It is visually obvious that projection can change lengths and even the ratio of lengths, because equal lengths often appear unequal under projection. And yet we can recognize that Figure 5.1 is a picture of equal tiles, even though they are unequal in size and shape. Some clue to their equality must be preserved, but what? It cannot be length; it cannot be a ratio of lengths; but, surprisingly, it can be a ratio of ratios, called the cross-ratio. The cross-ratio is a quantity associated with four points on a line. If the four points have coordinates p, q, r, and s, then their cross-ratio is the function of the ordered 4-tuple (p, q, r, s) defined by (r − p)/(s − p) , (r − q)/(s − q)
which can also be written as
(r − p)(s − q) . (r − q)(s − p)
The cross-ratio is preserved by projection. To show this, it suffices to show that it is preserved by the three generating transformations from which we composed all linear fractional maps in the previous section: 1. The map sending x to x + l. Here the numbers p, q, r, s are replaced by p + l, q + l, r + l, s + l, respectively. This does not change the cross-ratio because the l terms cancel by subtraction. 2. The map sending x to kx. Here the numbers p, q, r, s are replaced by kp, kq, kr, ks, respectively. This does not change the cross-ratio because the k terms cancel by division. 3. The map sending x to 1/x. Here the numbers p, q, r, s are replaced by 1p , 1q , 1r , 1s , respectively, so the cross-ratio (r − p)(s − q) (r − q)(s − p)
5.7 The cross-ratio
109
is replaced by ( 1r − 1p )( 1s − 1q ) ( 1r − 1q )( 1s − 1p )
=
p−r pr · q−r qr ·
q−s qs p−s ps
taking common denominators,
(p − r)(q − s) (q − r)(p − s) (r − p)(s − q) = (r − q)(s − p) =
multiplying through by pqrs, changing the sign in all factors,
and thus, the cross-ratio is unchanged in this case also.
Is the cross-ratio visible? If we take the four equally spaced points p = 0, q = 1, r = 2, and s = 3 on the line, then their cross-ratio is (r − p)(s − q) 2 × 2 4 = = . (r − q)(s − p) 1 × 3 3 It follows that any projective image of these points also has cross-ratio 4/3. Do four points on a line look equally spaced if their cross-ratio is 4/3? Test your eye on the quadruples of points in Figure 5.18, and then do Exercise 5.7.2 to find the correct answer.
Figure 5.18: Which is a projective image of equally spaced points?
Exercises We will see in the next section that any three points on RP1 can be projected to any three points. Hence, there cannot be an invariant involving just three points. However, the invariance of the cross-ratio tells us that, once the images of three points are known, the whole projection map is known (compare with the “three-point determination” of isometries of the plane in Section 3.7).
110
5 Perspective
5.7.1 Show that there is only one point s that has a given cross-ratio with given points p, q, and r. In particular, if we have the points p = 0, q = 2, r = 3 (which we do in the three quadruples in Figure 5.18), there is exactly one s that gives the cross-ratio 4/3 required for “equally spaced” points. 5.7.2 Find the value of s that gives the cross-ratio 4/3, and hence find the “equally spaced” quadruple in Figure 5.18. Before the discovery of perspective, artists sometimes attempted to draw a tiled floor by making the width of each row of tiles a constant fraction e of the one before. 5.7.3 Show that this method is not correct by computing the cross-ratio of four points separated by the distances 1, e, and e2 .
5.8 What is special about the cross-ratio? In the remainder of this book, we use the abbreviation [p, q; r, s] =
(r − p)(s − q) (r − q)(s − p)
for the cross-ratio of the four points p, q, r, s, taken in that order. We have shown that the cross-ratio is an invariant of linear fractional transformations, but it is obviously not the only one. Examples of other invariants are (cross-ratio)2 and cross-ratio +1. The cross-ratio is special because it is the defining invariant of linear fractional transformations. That is, the linear fractional transformations are precisely the transformations of RP1 that preserve the cross-ratio. (Thus, the cross-ratio defines linear fractional transformations the way that length defines isometries.) We prove this fact among several others about linear fractional transformations and the cross-ratio. Fourth point determination. Given any three points p, q, r ∈ RP1 , any other point x ∈ RP1 is uniquely determined by its cross-ratio [p, q; r, x] = y with p, q, r. This statement holds because we can solve the equation y=
(r − p)(x − q) (r − q)(x − p)
uniquely for x in terms of p, q, r, and y.
5.8 What is special about the cross-ratio?
111
Existence of three-point maps. Given three points p, q, r ∈ RP1 and three points p , q , r ∈ RP1 , there is a linear fractional transformation f sending p, q, r to p , q , r , respectively. This statement holds because there is a projection sending any three points p, q, r to any three points p , q , r , and any projection is linear fractional by Section 5.6. The way to project is shown in Figure 5.19. P p
q
r
p q r Figure 5.19: Projecting three points to three points Without loss of generality, we can place the two copies of RP 1 so that p = p . Then the required projection is from the point P where the lines qq and rr meet. Uniqueness of three-point maps. Exactly one linear fractional function sends three points p, q, r to three points p , q , r , respectively. A linear fractional f sending p, q, r to p , q , r , respectively, must send any x = p, q, r to x satisfying [p, q; r, x] = [p , q ; r , x ], because f preserves the cross-ratio by Section 5.7. But x is unique by fourth point determination, so there is exactly one such function f . Characterization of linear fractional maps. These are precisely the maps of RP1 that preserve the cross-ratio. By Section 5.7, any linear fractional map f preserves the cross-ratio. That is, [ f (p, f (q); f (r), f (s)] = [p, q; r, s] for any four points p, q, r, s. Conversely, suppose that f is a map of RP1 with [ f (p), f (q); f (r), f (s)] = [p, q; r, s] for any four points p, q, r, s. By the existence of three-point maps, we can find a linear fractional g that agrees with f on p, q, r. But then, because f preserves the cross-ratio, g agrees with f on s also, by unique fourth point determination. Thus, g agrees with f everywhere, so f is a linear fractional map.
112
5 Perspective
The existence of three-point maps says that any three points on RP 1 can be sent to any three points by a linear fractional transformation. Thus, any invariant of triples of points must have the same value for any triple, and so it is trivial. A nontrivial invariant must involve at least four points, and the cross-ratio is an example. It is in fact the fundamental example, in the following sense. The fundamental invariant. Any invariant of four points is a function of the cross-ratio. To see why, suppose I(p, q, r, s) is a function, defined on quadruples of distinct points, that is invariant under linear fractional transformations. Thus, I ( f (p), f (q), f (r), f (s)) = I(p, q, r, s) for any linear fractional f . In other words, I has the same value on all quadruples (p , q , r , s ) that result from (p, q, r, s) by a linear fractional transformation. But more is true: I has the same value on all quadruples (p , q , r , s ) with the same cross-ratio as (p, q, r, s), because such a quadruple (p , q , r , s ) results from (p, q, r, s) by a linear fractional transformation. This follows from the existence and uniqueness of three-point maps: • by existence, we can send p, q, r to p , q , r , respectively, by a linear fractional transformation f , and • by uniqueness, f also sends s to s , the unique point that makes [p, q; r, s] = [p , q ; r , s ]. Because I has the same value on all quadruples with the same cross-ratio, it is meaningful to view I as a function J of the cross-ratio, defined by J([p, q; r, s]) = I(p, q, r, s).
Exercises The following exercises illustrate the result above about invariant functions of quadruples. They show that the invariants obtained by permuting the variables in the cross-ratio y = [p, q; r, s] are simple functions of y, such as 1/y and y − 1. 5.8.1 If [p, q; r, s] = y, show that [p, q; s, r] = 1/y. 5.8.2 If [p, q; r, s] = y, show that [q, p; r, s] = 1/y.
5.9 Discussion
113
5.8.3 Prove that [p, q; r, s] + [p, r; q, s] = 1, so it follows that if [p, q; r, s] = y, then [p, r; q, s] = 1 − y. The transformations y → 1/y and y → 1 − y obtained in this way generate all transformations of the cross-ratio obtained by permuting its variables. There are six such transformations (even though there are 24 permutations of four variables). 5.8.4 Show that the functions of y obtained by combining 1/y and 1 − y in all ways are 1 1 1 y y, , 1 − y, 1 − , , . y y 1−y y−1 5.8.5 Explain why any permutation of four variables may be obtained by exchanges: either of the first two, the middle two, or the last two variables. 5.8.6 Deduce from Exercises 5.8.1–5.8.5 that the invariants obtained from the cross-ratio y by permuting its variables are precisely the six listed in Exercise 5.8.4. The six linear fractional functions of y obtained in Exercise 5.8.4 constitute what is sometimes called the cross-ratio group. It is an example of a concept we will study in Chapter 7: the concept of a group of transformations. Unlike most of the groups studied there, this group is finite.
5.9 Discussion The plane RP2 studied in this chapter is the most important projective plane, but it is far from being the only one. Many other projective planes can be constructed by imitating the construction of RP2 , which is based on ordered triples (x, y, z) and linear equations ax + by + cz = 0. It is not essential for x, y, z to be real numbers. As noted earlier, they could be complex numbers, but more generally they could be elements of any field. A field is any set with + and × operations satisfying the nine field axioms listed in Section 4.8. If F is any field, we can consider the space F3 of ordered triples (x, y, z) with x, y, z ∈ F. Then the projective plane FP2 has • “points,” each of which is a set of triples (kx, ky, kz), where x, y, z ∈ F are fixed and k runs through the elements of F, • “lines,” each of which consists of the “points” satisfying an equation of the form ax + by + cz = 0 for some fixed a, b, c ∈ F.
114
5 Perspective
The projective plane axioms can be checked for FP2 just as they were for RP2 . The same calculations apply, because the field axioms ensure that the same algebraic operations work in F (solving equations, for example). This gives a great variety of planes FP2 , because there are a great variety of fields F. Perhaps the most familiar field, after R and C, is the set Q of rational numbers. QP2 is not unlike RP2 , except that all of its points have rational coordinates, and all of its lines are full of gaps, because they contain only rational points. More surprising examples arise from taking F to be a finite field, of which there is one with pn elements for each power pn of each prime p. The simplest example is the field F2 , whose members are the elements 0 and 1, with the following addition and multiplication tables. + 0 1
0 0 1
× 0 1
1 1 0
0 0 0
1 0 1
The projective plane F2 P2 has seven points, corresponding to the seven nonzero points in F32 : (1, 0, 0), (0, 1, 0), (0, 0, 1), (0, 1, 1), (1, 0, 1), (1, 1, 0), (1, 1, 1). These points are arranged in threes along the seven lines in Figure 5.20, one of which is drawn as a circle so as to connect its three points. (0,0,1)
(1,0,1)
(0,1,1) (1,1,1)
(1,0,0)
(1,1,0)
(0,1,0)
Figure 5.20: The smallest projective plane
5.9 Discussion
115
Notice that the lines satisfy the linear equations x = 0, x + y = 0,
y = 0,
z = 0,
y + z = 0,
z + x = 0,
x + y + z = 0. For example, the points on the circle satisfy x + y + z = 0. (Of course, the coordinates have nothing to do with position in the plane of the diagram. The figure is mainly symbolic, while attempting to show “points” collected into “lines.”) This structure is called the Fano plane, and it is the smallest projective plane. Despite being small, it is well-behaved, because its “lines” satisfy linear equations, just as lines do in the traditional geometric world. Thanks to finite fields, linear algebra works well in many finite structures. It has led to the wholesale development of finite geometries, many of which have applications in the mathematics of information and communication. However, the three axioms for a finite projective plane do not ensure that the plane is of the form FP2 , with coordinates for points and linear equations for lines. They can be satisfied by bizarre “nonlinear” structures, as we will see in the next chapter. A fourth axiom is needed to engender a field F of coordinates, and the axiom is none other than the theorem of Pappus that we met briefly in Chapters 1 and 4. This state of affairs will be explained in Chapter 6.
The invariance of the cross-ratio The invariance of the cross-ratio was discovered by Pappus around 300 CE and rediscovered by Desargues around 1640. It appears (not very clearly) as Proposition 129 in Book VII of Pappus’ Mathematical Collection and again in Mani`ere universelle de Mr Desargues in 1648. The latter is a pamphlet on perspective by written by Abraham Bosse, a disciple of Desargues. It also contains the first published statement of the Desargues theorem mentioned in Chapters 1 and 4. Because of this, and the fact that he wrote the first book on projective geometry, Desargues is considered to be the founder of the subject. Nevertheless, projective geometry was little known until the 19th century, when geometry expanded in all directions. In the more general 19th century geometry (which often included use of complex numbers), the cross-ratio continued to be a central concept.
116
5 Perspective
One of the reasons we now consider the appropriate generalization of classical projective geometry to be projective geometry with coordinates in a field is that the cross-ratio continues to make sense in this setting. Linear fractional transformations and the cross-ratio make sense when R is replaced by any field F. The transformations x → x + l and x → kx make sense on F, and x → 1/x makes sense on F ∪ {∞} if we set 1/0 = ∞ and 1/∞ = 0. Then the transformations x →
ax + b , cx + d
where a, b, c, d ∈ F and ad − bc = 0,
make sense on the “F projective line” FP1 = F ∪ {∞}. The cross-ratio is invariant by the same calculation as in Section 5.7, thanks to the field axioms, because the usual calculations with fractions are valid in a field.
6 Projective planes P REVIEW In this chapter, geometry fights back against the forces of arithmetization. We show that coordinates need not be brought into geometry from outside—they can be defined by purely geometric means. Moreover, the geometry required to define coordinates and their arithmetic is simpler than Euclid’s geometry. It is the projective geometry introduced in the previous chapter, but we have to build it from scratch using properties of straight lines alone. We started this project in Section 5.3 by stating the three axioms for a projective plane. However, these axioms are satisfied by many structures, some of which have no reasonable system of coordinates. To build coordinates, we need at least one additional axiom, but for convenience we take two: the Pappus and Desargues properties that were proved with the help of coordinates in Chapter 4. Here we proceed in the direction opposite to Chapter 4: Take Pappus and Desargues as axioms, and use them to define coordinates. The coordinates are points on a projective line, and we add and multiply them by constructions like those in Chapter 1. But instead of using parallel lines as we did there, we call lines “parallel” if they meet on a designated line called the “horizon” or the “line at infinity.” The main problem is to prove that our addition and multiplication operations satisfy the field axioms. This is where the theorems of Pappus and Desargues are crucial. Pappus is needed to prove the commutative law of multiplication, ab = ba, whereas Desargues is needed to prove the associative law, a(bc) = (ab)c.
117
118
6 Projective planes
6.1 Pappus and Desargues revisited The theorems of Pappus and Desargues stated in Chapters 1 and 4 had a similar form: If two particular pairs of lines are parallel, then a third pair is parallel. Because parallel lines meet on the horizon, the Pappus and Desargues theorems also say that if two particular pairs of lines meet on the horizon, then so does a third pair. And because the horizon is not different from any other line, these theorems are really about three pairs of lines having their intersections on the same line. In this projective setting, the Pappus theorem takes the form shown in Figure 6.1. The six vertices of the hexagon are shown as dots, and the opposite sides are shown as a black pair, a gray pair, and a dotted pair. The line on which each of the three pairs meet is labeled L , and we have oriented the figure so that L is horizontal (but this is not at all necessary). Projective Pappus theorem. Six points, lying alternately on two straight lines, form a hexagon whose three pairs of opposite sides meet on a line. L
Figure 6.1: The projective Pappus configuration This statement of the Pappus theorem is called projective because it involves only the concepts of points, lines, and meetings between them. Meetings between geometric objects are called incidences, and, for this reason, the Pappus theorem is also called an incidence theorem. The three axioms of a projective plane, given in Section 5.3, are the simplest examples of incidence theorems.
6.1 Pappus and Desargues revisited
119
The projective Desargues theorem is another incidence theorem. It concerns the pairs of corresponding sides of two triangles, shown in solid gray in Figure 6.2. The triangles are in perspective from a point P, which means that each pair of corresponding vertices lies on a line through P. The three corresponding pairs of sides are again shown as black, gray, and dotted, and each pair meets on a line labeled L . Projective Desargues theorem. If two triangles are in perspective from a point, then their pairs of corresponding sides meet on a line. L
P
Figure 6.2: The projective Desargues configuration An important special case of the Desargues theorem has the center of projection P on the line L where the corresponding sides of the triangles meet. This special case is called the little Desargues theorem, and it is shown in Figure 6.3. Little Desargues theorem. If two triangles are in perspective from a point P, and if two pairs of corresponding sides meet on a line L through P, then the third pair of corresponding sides also meets on L . Because the projective Pappus and Desargues theorems involve only incidence concepts, one would like proofs of them that involve only the three axioms for a projective plane given in Section 5.3. Unfortunately, this is not possible, because there are examples of projective planes not satisfying the Pappus and Desargues theorems. What we can do, however, is take the Pappus and Desargues theorems as new axioms. Together with the original three axioms for projective planes, these two new axioms apply to a broad class of projective planes called Pappian planes.
120
6 Projective planes L
P
Figure 6.3: The little Desargues configuration The Pappian planes include RP2 and many other planes, but not all. They turn out to be the planes with coordinates satisfying the same laws of algebra as the real numbers—the field axioms. The object of this chapter is to show how coordinates arise when the Pappus and Desargues theorems hold, and why they satisfy the field axioms. In doing so, we will see that projective geometry is simpler than algebra in a certain sense, because we use only five geometric axioms to derive the nine field axioms.
Exercises In some projective planes, the Desargues theorem is false. Here is one example, which is called the Moulton plane. Its “points” are ordinary points of R2 , together with a point at infinity for each family of parallel “lines.” However, the “lines” of the Moulton plane are not all ordinary lines. They include the ordinary lines of negative, horizontal, or vertical slope, but each other “line” is a broken line consisting of a half line of slope k > 0 below the x-axis, joined to a half line of slope k/2 above the x-axis. Figure 6.4 shows some of the “lines.”
y
O
x
Figure 6.4: Lines of the Moulton plane
6.2 Coincidences
121
6.1.1 Find where the “line” from (0, −1) to (2, 1/2) meets the x-axis. 6.1.2 Explain why any two “points” of the Moulton plane lie on a unique “line.” 6.1.3 Explain why any two “lines” of the Moulton plane meet in a unique “point.” (Parallel “lines” have a common “point at infinity” by definition, so do not worry about them.) 6.1.4 Give four “points,” no three of which lie on the same “line.” 6.1.5 Thus, the Moulton plane satisfies the three axioms of a projective plane. But it does not satisfy even the little Desargues theorem, as Figure 6.5 shows. Explain.
y L O
x
Figure 6.5: Failure of the little Desargues theorem in the Moulton plane
6.2 Coincidences Two points A, B always lie on a line. But it is accidental, so to speak, if a third point C lies on the line through A and B. Such an accidental meeting is called a “coincidence” in everyday life, and this is a good name for it in projective geometry too: coincidence = two incidences together—in this case the incidence of A and B with a line, and the incidence of C with the same line. The theorems of Pappus and Desargues state that certain coincidences occur. In fact, they are coincidences of the type just described, in which two points lie on a line and a third point lies on the same line. The perspective picture of the tiled floor also involves certain coincidences, as becomes clear when we look again at the first few steps in its construction (Figure 6.6).
122
6 Projective planes
Draw diagonal of first tile, extended to the horizon
Extend diagonal of second tile to the horizon
Draw side of second tile, through the new intersection
Draw second column of tiles, through the new intersection
Figure 6.6: Constructing the tiled floor
At this step, a coincidence occurs. Three of the points we have constructed lie on a straight line, which is shown dashed in Figure 6.7.
Figure 6.7: A coincidence in the tiled floor
6.2 Coincidences
123
This coincidence can be traced to a special case of the little Desargues theorem, which involves the two shaded triangles shown in Figure 6.8. P
Figure 6.8: A little Desargues configuration in the tiled floor This case of little Desargues says that the two dotted lines (diagonals of “double tiles”) meet on the horizon. These lines give us a second little Desargues configuration, shown in Figure 6.9, from which we conclude that the dashed diagonals also meet on the horizon, as required to explain the coincidence in Figure 6.7. P
Figure 6.9: A second little Desargues configuration
Exercises The occurrence of the little Desargues configuration in the tiled floor may be easier to see if we draw the lines meeting on the horizon as actual parallels. The little Desargues theorem itself is easier to state in terms of actual parallels (Figure 6.10). 6.2.1 Formulate an appropriate statement of the little Desargues theorem when one has parallels instead of lines meeting on L .
124
6 Projective planes
Figure 6.10: The parallel little Desargues configuration 6.2.2 Now redraw Figures 6.7, 6.8, and 6.9 so that the lines meeting on L are shown as actual parallels. 6.2.3 What is the nature of the “coincidence” in Figure 6.7 now? 6.2.4 Find occurrences of the little Desargues configuration in your diagrams. Hence, explain why the “coincidence” in Exercise 6.2.3 follows from your statement of the little Desargues theorem in Exercise 6.2.1. The theorem that proves the coincidence in the drawing of the tiled floor is actually a special case of the little Desargues theorem: the case in which a vertex of one triangle lies on a side of the other. Thus, it is not clear that the coincidence is false in the Moulton plane, where we know only that the general little Desargues theorem is false by the exercises in Section 6.1. 6.2.5 By placing an x-axis in a suitable position on Figure 6.11, show that the tiled floor coincidence fails in the Moulton plane.
Figure 6.11: A coincidence that fails in the Moulton plane
6.3 Variations on the Desargues theorem
125
6.3 Variations on the Desargues theorem In Section 6.1, we stated the Desargues theorem in the form: If two triangles are in perspective from a point, then their three pairs of corresponding sides meet on a line. The Desargues theorem is a very flexible theorem, which appears in many forms, and two that we need later are the following. (We need these theorems only as consequences of the Desargues theorem, but they are actually equivalent to it.) Converse Desargues theorem. If corresponding sides of two triangles meet on a line, then the two triangles are in perspective from a point. To deduce this result from the Desargues theorem, let ABC and A BC be two triangles whose corresponding sides meet on the line L . Let P be the intersection of AA and BB , so we want to prove that P lies on CC as well. Suppose that PC meets the line BC at C (Figure 6.12 shows C , hypothetically, unequal to C ). L
Q
A
A P
B B C C C Figure 6.12: The converse Desargues theorem
Then the triangles ABC and A BC are in perspective from P and therefore, by the Desargues theorem, their corresponding sides meet on a line. We already know that AB meets A B on L , and that BC meets BC on L . Hence, AC meets AC on L , necessarily at the point Q where AC meets L . It follows that QA goes through C . But we also know that QA meets BC at C . Hence, C = C . Thus, C is indeed on the line PC, so ABC and A BC are in perspective from P, as required.
126
6 Projective planes
The second consequence of the Desargues theorem is called the “scissors theorem.” I do not know how common this name is, but it is used on p. 69 of the book Fundamentals of Mathematics II. Geometry, edited by Behnke, Bachmann, Fladt, and Kunle. In any case, it is an apt name, as you will see from Figure 6.13. Scissors theorem. If ABCD and A BC D are quadrilaterals with vertices alternately on two lines, and if AB is parallel to A B , BC to BC , and AD to A D , then also CD is parallel to C D . C
A C E
A E P
B
D
B
D
Figure 6.13: The scissors theorem To prove this theorem, let E be the intersection of AD and BC and let be the intersection of A D and BC , as shown in Figure 6.13. Then the triangles ABE and A B E have corresponding sides parallel. Hence, they are in perspective from the intersection P of AA and BB , by the converse Desargues theorem. But then the triangles CDE and C D E are also in perspective from P. Because their sides CE and C E , DE and D E , are parallel by assumption, it follows from the Desargues theorem that CD and C D are also parallel, as required. E
The scissors theorem just proved says that if the black, gray, and dashed lines in Figure 6.13 are parallel, then so are the dotted lines. What if the black, gray, and dotted lines are parallel: Are the dashed lines again parallel? The answer is yes, and the proof is similar, but with a slightly different picture (Figure 6.14).
6.3 Variations on the Desargues theorem
127
P
Figure 6.14: Second case of the scissors theorem We have extended the black and dotted lines until they meet, forming triangles with their corresponding black, gray, and dotted sides parallel. Then it follows from the converse Desargues theorem that these triangles are in perspective from P. But then so are the triangles with dashed, black, and dotted sides. Hence their dashed sides are parallel by the Desargues theorem. Remark. In practice, the scissors theorem is often used in the following way. We have a pair of scissors ABCD and another figure D A BC F with parallel pairs of black, gray, dashed, and dotted lines as shown in Figure 6.15. We want to prove that D = F (so the ends of the gray and dotted lines coincide, and the second figure is also a pair of scissors). C
A C A
P
B
D
B
D F
Figure 6.15: Applying the scissors theorem This coincidence happens because the line C D is parallel to CD by the scissors theorem, so C D is the same line as C F , and hence D = F .
128
6 Projective planes
Exercises Because the Desargues theorem implies its converse, another way to show that the Desargues theorem fails in the Moulton plane is to show that its converse fails. This plan is easily implemented with the help of Figure 6.16. (Moulton himself used this figure when he introduced the Moulton plane in 1902.)
y
P
O
x
Figure 6.16: The converse Desargues theorem fails in the Moulton plane 6.3.1 Explain why Figure 6.16 shows the failure of the converse Desargues theorem in the Moulton plane. 6.3.2 Formulate a converse to the little Desargues theorem, and show that it follows from the little Desargues theorem. 6.3.3 Show that the converse little Desargues theorem implies a “little scissors theorem” in which the quadrilaterals have their vertices on parallel lines. 6.3.4 Design a figure that directly shows the failure of the little scissors theorem in the Moulton plane.
6.4 Projective arithmetic If we choose any two lines in a projective plane as the x- and y-axes, we can add and multiply any points on the x-axis by certain constructions. The constructions resemble constructions of Euclidean geometry, but they use straightedge only, so they make sense in projective geometry. To keep them simple, we use lines we call “parallel,” but this merely means lines meeting on a designated “line at infinity.” The real difficulty is that the construction of a + b, for example, is different from the construction of b + a, so it is a “coincidence” if a + b = b + a. Similarly, it is a “coincidence” if ab = ba, or if any other law of algebra holds. Fortunately, we can show that the required coincidences actually occur, because they are implied by certain geometric coincidences, namely, the Pappus and Desargues theorems.
6.4 Projective arithmetic
129
Addition To construct the sum a + b of points a and b on the x-axis, we take any line L parallel to the x-axis and construct the lines shown in Figure 6.17: 1. A line from a to the point where L meets the y-axis. 2. A line from b parallel to the y-axis. 3. A parallel to the first line through the intersection of the second line and L . y L
a b
O
a+b
x
Figure 6.17: Construction of the sum This construction is similar in spirit to the construction of the sum in Section 1.1. There we “copied a length” by moving it from one place to another by a compass. The spirit of the compass remains in the projective construction: the black line and the gray line form a “compass” that “copies” the length Oa to the point b. We need the line L to construct a + b, but we get the same point a + b from any other line L parallel to the x-axis. This coincidence follows from the little Desargues theorem as shown in Figure 6.18. y L L
O
a b
a+b
x
Figure 6.18: Why the sum is independent of the choice of L The black sides of the solid triangles are parallel by construction, as are the gray sides, one of which ends at the point a + b constructed from L . Then it follows from the little Desargues theorem that the dotted sides are also parallel, and one of them ends at the point a + b constructed from L . Hence, the same point a + b is constructed from both L and L .
130
6 Projective planes
Multiplication To construct the product ab of two points a and b on the x-axis, we first need to choose a point = O on the x-axis to be 1. We also choose a point = O to be the 1 on the y-axis. The point ab is constructed by drawing the black and gray lines from 1 and a on the x-axis to 1 on the y-axis, and then drawing their parallels as shown in Figure 6.19. This construction is the projective version of “multiplication by a” done in Section 1.4. y
1 O
1
a
b
x
ab
Figure 6.19: Construction of the product ab Choosing the 1 on the x-axis means choosing a unit of length on the xaxis, so the position of ab definitely depends on it. For example, ab = b if a = 1 but ab = b if a = 1. However, the position of ab does not depend on the choice of 1 on the y-axis, as the scissors theorem shows (Figure 6.20). y
1 1 O
1
a
b
ab
x
Figure 6.20: Why the product is independent of the 1 on the y-axis If we choose 1 instead of 1 to construct ab, the path from b to ab follows the dashed and the dotted line instead of the black and the gray line. But it ends in the same place, because the dotted line to ab is parallel to the dotted line to a, by the scissors theorem.
6.4 Projective arithmetic
131
Interchangeability of the axes Once we have chosen points called 1 on both the x- and y-axes, it is natural to let each point a on the x-axis correspond to the point on the y-axis obtained by drawing the line through a parallel to the line through the points 1 on both axes (Figure 6.21). y a 1 O
a
1
x
Figure 6.21: Corresponding points It is also natural to define sum and product on the y-axis by constructions like those on the x-axis. But then the question arises: Do the y-axis sum and product correspond to the x-axis sum and product? To show that sums correspond, we need to construct a + b on the xaxis, and then show that the corresponding point a + b on the y-axis is the y-axis sum of the y-axis a and b. Figure 6.22 shows how this construction is done. y M
a+b b a
O
L
a
b
a+b
x
Figure 6.22: Corresponding sums We construct a + b on the x-axis using the line L through a on the y-axis. That is, draw the line M through b on the x-axis and parallel to the y-axis, and then draw the line (dashed) from the intersection of M and L parallel to the line from a on the x-axis to the intersection of L and the y-axis. This dashed line meets the x-axis at a + b, and (because it is parallel to the line from a to a) it also meets the y-axis at a + b.
132
6 Projective planes
Now we construct a + b on the y-axis using the line M (as on the xaxis, the sum does not depend on line chosen, as long as it is parallel to the y-axis). That is, draw the line through b on the y-axis parallel to the x-axis, and then draw the line (dotted) parallel to the line from a on the y-axis to b on the x-axis (because this b is the intersection of the x-axis with M ). But then, as is clear from Figure 6.22, we have a Pappus configuration of gray, dashed, and dotted lines between the y-axis and M , hence the dotted line (leading to the y-axis sum) and the dashed line (leading to the point corresponding to the x-axis sum) end at the same point, as required. To show that products correspond, we use the scissors theorem from Section 6.3. Figure 6.23 shows the corresponding points 1, a, b, and ab on both axes. The gray lines construct ab on the x-axis, and the dotted lines construct ab on the y-axis. y ab b a 1 x O
1
a
b
ab
Figure 6.23: Construction of the product ab on both axes It follows from the scissors theorem that the dotted line on the right ends at the same point as the black line from ab on the x-axis parallel to the lines connecting the corresponding points a and the corresponding points b. Hence, the product of a and b on the y-axis (at the end of the dotted line) is indeed the point corresponding to ab on the x-axis.
Exercises These definitions of sum and product lead immediately to some of the simpler laws of algebra, namely, those concerned with the behavior of 0 and 1. The complete list of algebraic laws is given in the Section 6.5.
6.5 The field axioms
133
6.4.1 Show that a + O = a for any a, so O functions as the zero on the x-axis. 6.4.2 Show that, for any a, there is a point b that serves as −a; that is, a + b = O. (Warning: Do not be tempted to use measurement to find b. Work backward from O = a + b, reversing the construction of the sum.) 6.4.3 Show that a1 = a for any a. 6.4.4 Show that, for any a = O, there is a b that serves as a−1 ; that is, ab = 1. (Again, do the construction of the product in reverse.) You will notice that we have not attempted to define sums or products involving the point at infinity ∞ on the x-axis. 6.4.5 What happens when we try to construct a + ∞? 6.4.6 What is −∞? You should find that the answers to Exercises 6.4.5 and 6.4.6 are incompatible with ordinary arithmetic. This is why we do not include ∞ among the points we add and multiply.
6.5 The field axioms In calculating with numbers, and particularly in calculating with symbols (“algebra”), we assume several things: that there are particular numbers 0 and 1; that each number a has a additive inverse, −a; that each number a = 0 has a reciprocal, a−1 ; and that the following field axioms hold. (We introduced these in the discussion of vector spaces in Section 4.8.) a + b = b + a,
ab = ba
a + (b + c) = (a + b) + c,
a(bc) = (ab)c
(commutative laws) (associative laws)
a + 0 = a,
a1 = a
(identity laws)
a + (−a) = 0,
aa−1 = 1
(inverse laws)
a(b + c) = ab + ac
(distributive law)
We generally use these laws unconsciously. They are used so often, and they are so obviously true of numbers, that we do not notice them. But for the projective sum and product of points, they are not obviously true. It is not even clear that a + b = b + a, because the construction of a + b is different from the construction of b + a. It is truly a coincidence that a + b = b + a in projective geometry, the result of a geometric coincidence of the type discussed in Section 6.2.
134
6 Projective planes
In this chapter, we show that just two coincidences—the theorems of Pappus and Desargues—imply all nine field axioms. In fact, it is known that Pappus alone is sufficient, because it implies Desargues. We do not prove this fact here, partly because it is difficult, and partly because the Desargues theorem itself is interesting: It implies all the field axioms except ab = ba. Thus, the theorems of Pappus and Desargues have algebraic content that can be measured accurately by the field axioms they imply. Pappus implies all nine, and Desargues only eight—all but ab = ba.
Proof of the commutative laws y
1 O
1
a
b
ba
x
Figure 6.24: Construction of the product ba We begin with the law ab = ba, which is the most important consequence of the Pappus theorem. Figure 6.24 shows the construction of ba from a and b, lying at the end of the second dotted line. It is different from the construction of ab, and Figure 6.25 shows the constructions of both ab and ba on the same diagram. y
1 O
1
a
b
ab = ba
x
Figure 6.25: Construction of both ab and ba Then ab = ba because the gray and dotted lines end at the same place, by the Pappus theorem. The Pappus configuration in Figure 6.25 consists of all the lines except the line joining 1 on the x-axis to 1 on the y-axis.
6.5 The field axioms
135
There is a similar proof that a + b = b + a. Remember from Section 6.4 that a + b is the result of attaching the segment Oa at b. Thus, b + a is the result of attaching Ob at a, which is different from the construction of a + b. Looking at both constructions together (Figure 6.26), we see that the gray line leads to a + b and the dotted line leads to b + a. However, both of these lines end at the same point, thanks to the Pappus theorem. y L
O
a
b
a+b = b+a
x
Figure 6.26: Construction of both a + b and b + a
Exercises 6.5.1 Look back to the vector proof of the Pappus theorem in Section 4.2, and point out where it uses the assumption ab = ba. The Pappus configuration that proves a + b = b + a is actually a special one, because the vertices of the hexagon lie on parallel lines. The same special configuration also occurs in the proof in Section 6.4 that sums correspond on the x- and y-axes. The special configuration corresponds to a special Pappus theorem, sometimes called the “little Pappus theorem.” It is usually stated without mention of parallel lines; in which case, one has to talk about opposite sides of the hexagon meeting on a line L . 6.5.2 Given that the assumptions of the little Pappus theorem are a hexagon with vertices on two lines that meet at a point P, and two pairs of opposite sides meeting on a line L that goes through P, what is the conclusion? It is known that the little Desargues theorem implies the little Pappus theorem; a proof is in Fundamentals of Mathematics, II by Behnke et al., p. 70. Thus, the results deduced here from the little Pappus theorem can also be deduced from the little Desargues theorem (although generally not as easily).
136
6 Projective planes
6.6 The associative laws First we look at the associative law of addition, a + (b + c) = (a + b) + c. Figure 6.27 shows the construction of a + (b + c). We have to construct b + c from b and c first, and then add a as was done in Figure 6.17. Next we have to construct (a + b) + c, which means constructing a + b first and then adding it to c as shown in Figure 6.28. y L O
a
c
b
b+c
a + (b + c)
x
Figure 6.27: Construction of a + (b + c) y L O
a
a+b c
b
(a + b) + c
x
Figure 6.28: Construction of (a + b) + c Figure 6.29 shows both Figures 6.27 and 6.28 on the same diagram. Here we need Desargues or, more precisely, the scissors theorem. y L O
a
b
a+b c
b+c
a + (b + c) = (a + b) + c
Figure 6.29: Why a + (b + c) and (a + b) + c coincide
x
6.6 The associative laws
137
One can clearly see two pairs of scissors, each consisting of a dashed line, a dotted line, a black line, and a gray line. In the scissors on the right, the gray line ends at a + (b + c) and the dotted line at (a + b) + c. But the ends of these lines coincide, by the scissors theorem. Hence a + (b + c) = (a + b) + c. Because the scissors in this proof lie between parallel lines, we need only the little scissors theorem (and hence only the little Desargues theorem, by the remark in the previous exercise set). Next we consider the associative law of multiplication, a(bc) = (ab)c. The diagram (Figure 6.30) is similar, except that the pairs of scissors lie between nonparallel lines (the x- and y-axes), so now we need the full Desargues theorem. y
O
1 a b
ab
c
bc
a(bc) = (ab)c
x
Figure 6.30: Why a(bc) and (ab)c coincide The gray line ends at a(bc) and the dotted line ends at (ab)c. But the ends of these lines coincide, by the scissors theorem, so a(bc) = (ab)c.
Exercises There is an algebraic system that satisfies all of the field axioms except the commutative law of multiplication. It is called the quaternions and is denoted by H, after Sir William Rowan Hamilton, who discovered the quaternions in 1843. In 1845, Arthur Cayley showed that the quaternions could be defined as 2 × 2 complex matrices of the form a + ib c + id q= . −c + id a − ib Most of their properties follow from general properties of matrices. In fact, all the laws of algebra are immediate except the existence of q−1 and commutative multiplication.
138
6 Projective planes
6.6.1 Show that q has determinant a2 + b2 + c2 + d 2 and hence that q−1 exists for any nonzero quaternion q. 6.6.2 Find specific quaternions s and t such that st = ts. We can write any quaternion as q = a1 + bi + cj + dk, where 1 0 i 0 0 1 0 i 1= , i= , j= , k= . 0 1 0 −i −1 0 i 0 6.6.3 Verify that i2 = j2 = k2 = ijk = −1 (This is Hamilton’s description of the quaternions). It is possible to define the quaternion projective plane HP2 using quaternion coordinates. HP2 satisfies the Desargues theorem because it is possible to do the necessary calculations without using commutative multiplication. But it does not satisfy the Pappus theorem, because this implies commutative multiplication for H. HP2 is therefore a non-Pappian plane—probably the most natural example.
6.7 The distributive law To prove the distributive law a(b + c) = ab + ac, we take advantage of the ability to do addition and multiplication on both axes. We construct b + c from b and c on the x-axis, and then map b, c, and b + c to ab, ac, and a(b + c) on the y-axis by lines parallel to the line from 1 on the x-axis to a on the y-axis. Then we use addition on the y-axis to construct ab + ac there, and finally, use the Pappus theorem to show that ab + ac and a(b + c) are the same point. Here are the details. First, observe that we can map any b on the x-axis to ab on the y-axis by sending it along a line parallel to the line from 1 on the x-axis to a on the y-axis (Figure 6.31). ab
y
b a 1 O
1
b
Figure 6.31: Multiplication via parallels
x
6.7 The distributive law
139
This is the same as constructing ab from a and b on the y-axis, because the line from b to b is parallel to the line from 1 to 1, as required by the definition of multiplication. Next we add b and c on the x-axis, using a special choice of line L : the parallel through ab on the y-axis. We also connect b, c, and b + c, respectively, to ab, ac, and a(b + c) on the y-axis by parallel lines, shown dashed in Figure 6.32. The line through c that constructs b + c, namely the parallel M to the y-axis, is used in turn to add ab and ac on the y-axis. y a(b + c)
ab + ac
M
ac ab
O
L
b
c
b+c
x
Figure 6.32: Why a(b + c) = ab + ac This figure has the same structure as Figure 6.22; only the labels have changed. Now the dashed line ends at a(b + c), and the dotted line ends at ab + ac. But again the endpoints coincide by the theorem of Pappus, and so a(b + c) = ab + ac.
Exercises We need not prove the other distributive law, (b + c)a = ba + bc, because we are assuming Pappus, so multiplication is commutative. 6.7.1 Explain in this case why a(b + c) = ab + ac implies (b + c)a = ba + bc. However, in some important systems with noncommutative multiplication, both distributive laws remain valid. 6.7.2 Explain why both distributive laws are valid for the quaternions. 6.7.3 More generally, show that both distributive laws are valid for matrices.
140
6 Projective planes
6.8 Discussion The idea of developing projective geometry without the use of numbers comes from the German mathematician Christian von Staudt in 1847. His compatriots Hermann Wiener and David Hilbert took the idea further in the 1890s, and it reached a high point with the publication of Hilbert’s book, Grundlagen der Geometrie (Foundations of geometry), in 1899. It was Hilbert who first established a clear correlation between geometric and algebraic structure: • Pappus with commutative multiplication, • Desargues with associative multiplication. The correlation is significant because some important algebraic systems satisfy all the field axioms except commutative multiplication. The bestknown example is the quaternions, which has been known since 1843, but, for some reason, Hilbert did not mention it. To construct a non-Pappian plane, he created a rather artificial noncommutative coordinate system. It is perhaps a lucky accident of history that Hilbert discovered the role of the Desargues theorem at all. He was forced to use it because, in 1899, it was still not known that Pappus implies Desargues. This implication was first proved by Gerhard Hessenberg in 1904. Even then the proof was faulty, and the mistake was not corrected until years later. The whole circle of ideas was neatly tied up by yet another German mathematician, Ruth Moufang, in 1930. She found that the little Desargues theorem also has algebraic significance. In a projective plane satisfying the little Desargues theorem, with addition and multiplication defined as in Section 6.4, one can prove all the field axioms except commutativity and associativity. One can even prove a partial associativity law called cancellation or alternativity: a−1 (ab) = b = (ba)a−1
(alternativity)
The commutative, associative, and alternative laws are beautifully exemplified by the possible multiplication operations that can be defined “reasonably” on the Euclidean spaces Rn . (“Reasonably” means respecting at least the dimension of the space. For more on the problem of generalizing the idea of number to n dimensions, see the book Numbers by D. Ebbinghaus et al.)
6.8 Discussion
141
• Commutative multiplication is possible only on R1 and R2 , and it yields the number systems R and C. • Associative, but noncommutative, multiplication is possible only on R4 , and it yields the quaternions H. • Alternative, but nonassociative, multiplication is possible only on R8 , and it yields a system called the octonions O. The octonions were discovered by a friend of Hamilton called John Graves, in 1843, and they were discovered independently by Cayley in 1845. Ruth Moufang was the first to recognize the importance of quaternions and octonions in projective geometry. She pointed out the quaternion projective plane, as a natural example of a non-Pappian plane, and was the first to discuss the octonion projective plane OP2 . OP2 is the most natural example of a plane that satisfies little Desargues but not Desargues. In Section 5.4, we sketched the construction of the real projective space 3 RP by means of homogeneous coordinates. This idea is easily generalized to obtain the n-dimensional real projective space RP n , and one can obtain CPn and HPn in precisely the same way. Surprisingly, the idea does not work for the octonions. The only octonion projective spaces are the octonion projective line OP1 = O ∪ {∞} and the octonion projective plane OP2 discovered by Moufang. The reason for the nonexistence of OP3 is extremely interesting and has to do with the nature of the Desargues theorem in three dimensions. Remember that the Desargues theorem assumes a pair of triangles in perspective and concludes that the intersections of corresponding sides lie on a line. We know (because of the example of the Moulton plane) that the conclusion does not follow by basic incidence properties of points and lines. But if the triangles lie in three-dimensional space, the conclusion follows by basic incidence properties of points, lines, and planes. The spatial Desargues theorem is clear from a picture that emphasizes the placement of the triangles in three dimensions, such as Figure 6.33. The planes containing the two triangles meet in a line L , where the pairs of corresponding sides necessarily meet also. The argument is a little trickier if the two triangles lie in the same plane. But, provided the plane lies in a projective space, it can be carried out (one shows that the planar configuration is a “shadow” of a spatial configuration).
142
6 Projective planes
L
Figure 6.33: The spatial Desargues configuration Thus, the Desargues theorem holds in any projective space of at least three dimensions. This is why OP3 cannot exist. If it did, the Desargues theorem would hold in it, and we could then show that O is associative— which it is not. Q. E. D.
7 Transformations P REVIEW In this book, we have seen at least three geometries: Euclidean, vector, and projective. In Euclidean geometry, the basic concept is length, but angle and straightness derive from it. In vector geometry, the basic concepts are vector sums and scalar multiples, but from these we derive others, such as midpoints of line segments. Finally, in projective geometry, the basic concept is straightness. Length and angle have no meaning, but a certain combination of lengths—the cross-ratio—is meaningful because it is unchanged by projection. We found the cross-ratio as an invariant of projective transformations. The concept of length was not discovered this way, but nevertheless, it is an invariant of certain transformations. It is an invariant of the isometries, for the simple reason that isometries are defined to be transformations of the plane that preserve length. These examples are two among many that suggest geometry is the study of invariants of groups of transformations. This definition of geometry was first proposed by the German mathematician Felix Klein in 1872. Klein’s concept of geometry is perhaps still not broad enough, but it does cover the geometry in this book. In this chapter we look again at Euclidean, vector, and projective geometry from Klein’s viewpoint, first explaining precisely what “transformation” and “group” mean. It turns out that the appropriate transformations for projective geometry are linear, and that linear transformations also play an important role in Euclidean and vector geometry. Linear transformations also pave the way for hyperbolic geometry, a new “non-Euclidean” geometry that we study in Chapter 8.
143
144
7 Transformations
7.1 The group of isometries of the plane In Chapter 3, we took up Euclid’s idea of “moving” geometric figures, and we made it precise in the concept of an isometry of the plane R 2 . An isometry is defined to be a function f : R2 → R2 that preserves distance; that is, | f (P1 ) f (P2 )| = |P1 P2 | for any points P1 , P2 , ∈ R2 , where |P1 P2 | = (x2 − x1 )2 + (y2 − y1 )2 denotes the distance between the points P1 = (x1 , y1 ) and P2 = (x2 , y2 ). It follows immediately from this definition that, when f and g are isometries, so is their composite or product f g (the result of applying g, then f ). Namely, | f (g(P1 )) f (g(P2 ))| = |g(P1 )g(P2 )| because f is an isometry = |P1 P2 | because g is an isometry. What is less obvious is that any isometry f has an inverse, f −1 , which is also an isometry. To prove this fact, we use the result from Section 3.7 that any isometry of R2 is the product of one, two, or three reflections. First suppose that f = r1 r2 r3 , where r1 , r2 , and r3 are reflections. Then, because a reflection composed with itself is the identity function, we find f r3 r2 r1 = r 1 r2 r3 r3 r2 r1 = r 1 r2 r2 r1 = r 1 r1
because r3 r3 is the identity function
because r2 r2 is the identity function
= identity function, and therefore, r3 r2 r1 = f −1 . This calculation also shows that f −1 is an isometry, because it is a product of reflections. The proof is similar (but shorter) when f is the product of one or two reflections. These properties of isometries are characteristic of a group of transformations. A transformation of a set S is a function from S to S, and a collection G of transformations forms a group if it has the two properties: • If f and g are in G, then so is f g. • If f is in G, then so is its inverse, f −1 . It follows that G includes the identity function f f −1 , which can be written as 1. This notation is natural when we write the composite of two functions f , g as the “product” f g.
7.1 The group of isometries of the plane
145
What is a geometry? In 1872, the German mathematician Felix Klein pointed out that various kinds of geometry go with various groups of transformations. For example, the Euclidean geometry of R2 goes with the group of isometries of R2 . The meaningful concepts of the geometry correspond to properties that are left unchanged by transformations in the group. Isometries of R 2 leave distance or length unchanged, so distance is a meaningful concept of Euclidean geometry. It is called an invariant of the isometry group of R2 . This invariance is no surprise, because isometries are defined as the transformations that preserve distance. However, it is interesting that other things are also invariant under isometries, such as straightness of lines and circularity of circles. It is not entirely obvious that a length-preserving transformation preserves straightness, but it can be proved by showing first that any reflection preserves straightness, and then using the theorem of Section 3.7, that any isometry is a product of reflections. An example of a concept without meaning in Euclidean geometry is “being vertical,” because a vertical line can be transformed to a nonvertical line by an isometry (for example, by a rotation). We can do without the concept of “vertical” in geometry because we have the concept of “being relatively vertical,” that is, perpendicular. A concept that is harder to do without is “clockwise order on the circle.” This concept has no meaning in Euclidean geometry because the points A = (−1, 0), B = (0, 1), C = (1, 0), and D = (0, −1) have clockwise order on the circle, but their respective reflections in the x-axis do not. However, we can define oriented Euclidean geometry, in which clockwise order is meaningful, by using a smaller group of transformations. Instead of the group Isom(R2 ) of all isometries of R2 , take Isom+ (R2 ), each member of which is the product of an even number of reflections. Isom+ (R2 ) is a group because • If f and g are products of an even number of reflections, so is f g. • If f = r1 r2 · · · r2n is the product of an even number of reflections, then so is f −1 . In fact, f −1 = r2n · · · r2 r1 , by the argument used above to invert the product of any number of reflections. And any transformation in Isom+ (R2 ) preserves clockwise order because any product of two reflections does: The first reflection reverses the order, and then the second restores it.
146
7 Transformations
This example shows how a geometry of R2 with more concepts comes from a group with fewer transformations. In R3 , one has the concept of “handedness”—which distinguishes the right hand from the left—which is not preserved by all isometries of R3 . However, it is preserved by products of an even number of reflections in planes. Thus, the geometry of Isom(R 3 ) does not have the concept of handedness, but the geometry of Isom + (R3 ) does. Restricting the transformations to those that preserve orientation—as it is generally called—is a common tactic in geometry. However, the main goal of this book is to show that there are interesting geometries with fewer concepts than Euclidean geometry. These geometries are obtained by taking larger groups of transformations, which we study in the remainder of this chapter.
Exercises 7.1.1 Use the results of Section 3.7 to show that each member of Isom+ (R2 ) is either a translation or a rotation. 7.1.2 Why does an isometry map any circle to a circle? We took care to write the inverse of the isometry r1 r2 r3 as r3 r2 r1 because only this ordering of terms will always give the correct result. 7.1.3 Give an example of two reflections r1 and r2 such that r1 r2 = r2 r1 .
7.2 Vector transformations In Chapter 4, we viewed the plane R2 as a real vector space, by considering its points to be vectors that can be added and multiplied by scalars. If u = (u1 , u2 ) and v = (v1 , v2 ), we defined the sum of u and v by u + v = (u1 + v1 , u2 + v2 ) and the scalar multiple au of u by a real number a by au = (au1 , au2 ). A transformation f of R2 preserves these two operations on vectors if f (u + v) = f (u) + f (v) and and such a transformation is called linear.
f (au) = a f (u),
(*)
7.2 Vector transformations
147
One reason for calling the transformation “linear” is that it preserves straightness of lines. A straight line is a set of points of the form a + tb, where a and b are constant vectors and t runs through the real numbers. Figure 7.1 shows the role of the vectors a and b: a is one point on the line, and b gives the direction of the line.
a + tb
a b 0 Figure 7.1: Points on a line If we apply a linear transformation f to this set of points, we get the set of points f (a + tb). And by the linearity conditions (*), this set consists of points of the form f (a) + t f (b), which is another straight line: f (a) is one point on it, and f (b) gives the direction of the line. It follows from this calculation that, if L1 and L2 are two lines with direction b and f is a linear transformation, then f (L1 ) and f (L2 ) are two lines with direction f (b). In other words, a linear transformation also preserves parallels.
Matrix representation Another consequence of the linearity conditions (*) is that each linear transformation f of R2 can be specified by four real numbers a, b, c, d: any point (x, y) of R2 is sent by f to the point (ax + by, cx + dy). Certainly, there are numbers a, b, c, d that give the particular values f ((1, 0)) = (a, c) and
f ((0, 1)) = (b, d).
But the value of f ((x, y)) follows from these particular values by linearity: (x, y) = x(1, 0) + y(0, 1),
148
7 Transformations
and therefore, f ((x, y)) = f (x(1, 0) + y(0, 1)) = x f ((1, 0)) + y f ((0, 1)) = x(a, c) + y(b, d) = (ax + by, cx + dy). The linear transformation (x, y) → (ax + by, cx + dy) is usually represented by the matrix M=
a b c d
,
where a, b, c, d ∈ R.
2 To find where (x, y) ∈ R is sent by f , one writes it as the “column vector” x and multiplies this column on the left by M according to the matrix y product rule: a b x ax + by = . c d y cx + dy
The main advantage of the matrix notation is that it gives the product of two linear transformations, first (x, y) → (a2 x + b2 y, c2 x + d2 y) and then (x, y) → (a1 x + b1 y, c1 x + d1 y), by the matrix product rule:
a1 b1 c1 d1
a2 b2 c2 d2
=
a1 a2 + b1 c2 a1 b2 + b1 d2 c1 a2 + d1 c2 c1 b2 + d1 d2
.
Matrix notation also exposes the role of the determinant, det(M), which must be nonzero for the linear transformation to have an inverse. If M=
a b c d
,
then
det(M) = ad − bc,
and if det(M) = 0, then M −1 =
1 det M
d −b −c a
.
7.2 Vector transformations
149
Examples of linear transformations Any 2 × 2 real matrix M represents a linear transformation, because it follows from the definition of matrix multiplication that M(u + v) = Mu + Mv and M(au) = aMu for any vectors u and v (written in column form). Among the invertible linear transformations are certain isometries, such as rotations and reflections in lines through the origin. Recall from Section 3.6 that a rotation is a transformation of the form c −s (x, y) → (cx−sy, sx+cy), hence given by the matrix R = . s c The numbers c and s satisfy c2 + s2 = 1 (they are actually cos θ and sin θ , where θ is the angle of rotation); hence, c s det R = 1 and therefore R−1 = . −s c Likewise, reflection in the x-axis is the linear transformation 1 0 (x, y) → (x, −y), given by the matrix X = . 0 −1 We can reflect R2 in any line L through O with the help of the rotation R that sends the x-axis to L : • First apply R−1 to send L to the x-axis. • Then carry out the reflection by applying X. • Then send the line of reflection back to L by applying R. In other words, to reflect the point u in L , we find the value of RXR −1 u. Hence, reflection in L is represented by the matrix RXR−1 . Thus, the linear transformations of R2 include the isometries that are products of reflection on lines through O. But this is not all. An example of a linear transformation that is not an isometry is the stretch by factor k in the x-direction, k 0 (x, y) → (kx, y), given by the matrix S = . 0 1 It can be shown that any invertible linear transformation of R 2 is a product of reflections in lines through O and stretches in the x-direction (by factors k = 0).
150
7 Transformations
Affine transformations Linear transformations preserve geometrically natural properties such as straightness and parallelism, but they also preserve the origin, which really is not geometrically different from any other point. To abolish the special position of the origin, we allow linear transformations to be composed with translations, obtaining what are called affine transformations. If we write an arbitrary linear transformation of (column) vectors u in the form f (u) = Mu,
where M is an invertible matrix,
then an arbitrary affine transformation takes the form g(u) = Mu + c,
where c is a constant vector.
Because translations preserve everything except position, affine transformations preserve everything that linear transformations do, except position. In effect, they allow any point to become the origin. The geometry of affine transformations is called affine geometry. Its theorems include those in the first few sections of Chapter 4, such as the fact that diagonals of a parallelogram bisect each other, and the concurrence of the medians of a triangle. These theorems belong to affine geometry because they are concerned only with quantities, such as the midpoint of a line segment, that are preserved by affine transformations.
Exercises 7.2.1 Compute
MM −1
for the general 2 × 2 matrix M = 1 0 that it equals the identity matrix . 0 1
a c
b d
, and verify
7.2.2 Write down the matrix for clockwise rotation through angle π /4. 7.2.3 Write down the matrix for reflection in the line y = x, and check that it equals RXR−1 , where R is the matrix for rotation through π /4 found in Exercise 7.2.2. k 0 7.2.4 The matrix M = represents a dilation of the plane by factor k 0 k (also known as a similarity transformation). Explain geometrically why this transformation is a product of reflections in lines through O and of stretches by factor k in the x-direction.
7.3 Transformations of the projective line
151
7.2.5 Show that the midpoint of any line segment is preserved by linear transformations and hence by affine transformations. 7.2.6 More generally, show that the ratio of lengths of any two segments of the same line is preserved by affine transformations.
7.3 Transformations of the projective line Looking back at our approach to the projective line in Chapter 5, we see that we were following Klein’s idea. First we found the transformations of the projective line, and then a quantity that they leave invariant—the crossratio. In this section we look more closely at projective transformations, and show that they too can be viewed as linear transformations. In Sections 5.5 and 5.6, we showed that the transformations of the projective line R ∪ {∞} are precisely the linear fractional functions f (x) =
ax + b cx + d
where ad − bc = 0.
We did this by showing: • Any linear fractional function is a product of functions sending x to x + l, x to kx, and x to 1/x, and that each of the latter functions can be realized by projection of one line onto another. • Conversely, any projection of one line onto another is represented by a linear fractional function of x, with the understanding that 1/0 = ∞ and 1/∞ = 0. In the exercises to Section 5.6 you were asked to showthat linear fractional a b functions f (x) = ax+b , by showcx+d behave like the matrices M = c d ing that composition of the functions corresponds to multiplication of the corresponding matrices. In this section, we explain the connection by representing mappings of the projective line directly by linear transformations of the plane. We begin by defining the projective line in the manner of Section 5.4. There we defined the real projective plane RP2 . Its “points” are the lines through O in R3 , and its “lines” are the planes through O. Here we need only one projective line, which we can take to be the real projective line RP1 , whose “points” are the lines through O in the ordinary plane R 2 .
152
7 Transformations
We label each line through O, if it meets the line y = 1, by the xcoordinate s of the point of intersection (Figure 7.2). The single line that does not meet y = 1, namely, the x-axis, naturally gets the label ∞. y s s
1
O
x
Figure 7.2: Correspondence between lines through O and points on y = 1 Figure 7.3 shows some lines through O with their labels. 0 −1 −2 −3
1 2 3 ∞
Figure 7.3: Labeling of lines through O Now a projective map of the ordinary line y = 1 sends the point with xcoordinate s to the point with x-coordinate f (s), for some linear fractional function as + b . f (s) = cs + d
7.3 Transformations of the projective line
153
This function corresponds to a map of the plane R2 sending the line with label s to the line with label as+b cs+d . Bearing in mind that the label represents the “reciprocal slope” (“run over rise”) of the line, we find that one such map is the linear map of the plane given by the matrix M=
a b c d
.
To see why, we apply this linear map to a typical point (sx, x) on the line with label s. We find where M sends it by writing (sx, x) as a column vector and multiplying it on the left by M:
a b c d
sx x
=
asx + bx csx + dx
.
asx + bx represents the point (asx + bx, csx + dx), csx + dx which lies on the line with reciprocal slope
The column vector
asx + bx as + b = . csx + dx cs + d The latter line is therefore independent of x and it is the line with label as+b as+b cs+d . Thus, M maps the line with label s to the line with label cs+d , as required. Because a “point” of RP1 is a whole line through O, we care only that the matrix a b M= c d sends the line with label s to the line with label as+b cs+d . It does not matter how M moves individual points on the line. It is not generally the case that M sends the particular point (s, 1) on the line with label s to the particular as+b point ( as+b cs+d , 1) on the line with label cs+d . Indeed, it is clearly impossible when M represents the map s → 1/s of RP1 . A matrix M sends each point of R2 to another point of R2 . Hence it cannot send (0, 1) to (1/0, 1) = (∞, 1), because the latter is not a point of R2 . However, M can send the line with label 0 (the y-axis) to the line with label ∞ (the x-axis), and this is exactly what happens (see Exercises 7.3.1 and 7.3.2).
154
7 Transformations
It should also be pointed out that the representation of linear fractional functions by matrices is not unique. The fraction as + b cs + d
is equal to
Hence, the function f (s) = M=
a b c d
as+b cs+d
kas + kb kcs + kd
for any k = 0.
is represented not only by
but also by
kM =
ka kb kc kd
for any k = 0.
The linear transformations kM combine the transformation M with dilation by factor k, so they are all different. Thus, the same linear fractional function f is represented by infinitely many different transformations kM of R 2 . The message, again, is that we care only that each of these transformations sends the line with label s to the line with label f (s).
Exercises 7.3.1 Write down a matrix M that represents the map s → 1/s of RP1 . 7.3.2 Verify that your matrix M in Exercise 7.3.1 maps the y-axis onto the x-axis. 7.3.3 Sketch a picture of the lines with labels 1/2, −1/2, 1/3, and −1/3. The nonuniqueness of the matrix M corresponding to the linear fractional function f raises the question: Is there a natural way to choose one matrix for each linear fractional function? Actually, no, but there is a natural way to choose two matrices. 7.3.4 Given that
M=
a b c d
and ad − bc = 0,
show that the determinant of kM has absolute value 1 for exactly two of the matrices kM, where k = 0.
7.4 Spherical geometry The unit sphere in R3 consists of all points at unit distance from O, that is, all points (x, y, z) satisfying the equation x2 + y2 + z2 = 1.
7.4 Spherical geometry
155
This surface is also called the 2-sphere, or S2 , because its points can be described by two coordinates—latitude and longitude, for example. Its geometry is essentially two-dimensional, like that of the Euclidean plane R2 or the real projective plane RP2 , and indeed the fundamental objects of spherical geometry are “points” (ordinary points on the sphere) and “lines” (great circles on the sphere). However, like the projective plane RP2 , the sphere S2 is best understood via properties of the three-dimensional space R3 . In particular, the “lines” on S2 are the intersections of S2 with planes through O in R3 —the great circles—and the isometries of S2 are precisely the isometries of R3 that leave O fixed. By definition, an isometry f of R3 preserves distance. Hence, if f leaves O fixed, it sends each point at distance 1 from O to another point at distance 1 from O. In other words, an isometry f of R3 that fixes O also maps S2 into itself. The restriction of f to S2 is therefore an isometry of S2 , because f preserves distances on S2 as it does everywhere else. This statement is true whether one uses the straight-line distance between points of S2 or, as is more natural, the great-circle distance along the curved surface of S2 (Figure 7.4). The isometries of S2 are the maps of S2 into itself that preserve great circle distance, and we will see next why they are all restrictions of isometries of R3 .
O
1 Q P
θ
Figure 7.4: Great-circle distance
156
7 Transformations
The isometries of S2 The simplest isometries of R3 that fix O are reflections in planes through O. The corresponding isometries of S2 are the reflections in great circles. Two planes P1 and P2 meet in a line L through O, and the product of reflections in P1 and P2 is a rotation about L (through twice the angle between P1 and P2 ). This situation is completely analogous to that in R2 , where the product of reflections through O is a rotation (through twice the angle between the lines). Finally, there are products of reflections in three planes that are different from products of reflections in one or two planes. One such isometry is the antipodal map sending each point (x, y, z) to its antipodal point (−x, −y, −z). This map is the product of • reflection in the (y, z)-plane, which sends (x, y, z) to (−x, y, z), • reflection in the (z, x)-plane, which sends (x, y, z) to (x, −y, z), • reflection in the (x, y)-plane, which sends (x, y, z) to (x, y, −z). As in R2 , there is a “three reflections theorem” that any isometry of is the product of one, two, or three reflections. The proof is similar to the proof for R2 in Sections 3.3 and 3.7 (see the exercises below). This three reflections theorem shows why all isometries of S2 are restrictions of isometries of R3 , namely, because this is true of reflections in great circles. S2
Exercises The proof of the three reflections theorem begins, as it did for R2 , by considering the equidistant set of two points. 7.4.1 Show that the equidistant set of two points in R3 is a plane. Show also that the plane passes through O if the two points are both at distance 1 from O. 7.4.2 Deduce from Exercise 7.4.1 that the equidistant set of two points on S2 is a “line” (great circle) on S2 . Next, we establish that there is a unique point on S2 at given distances from three points not in a “line.” 7.4.3 Suppose that two points P, Q ∈ S2 have the same distances from three points A, B,C ∈ S2 not in a “line.” Deduce from Exercise 7.4.2 that P = Q. 7.4.4 Deduce from Exercise 7.4.3 that an isometry of S2 is determined by the images of three points A, B,C not in a “line.”
7.5 The rotation group of the sphere
157
Thus, it remains to show the following. Any three points A, B,C ∈ S2 not in a “line” can be mapped to any other three points A , B ,C ∈ S2 , which are separated by the same respective distances, by one, two, or three reflections. 7.4.5 Complete this proof of the three reflections theorem by imitating the argument in Section 3.7.
7.5 The rotation group of the sphere The group Isom(S2 ) of all isometries of S2 has a subgroup Isom+ (S2 ) consisting of the isometries that are products of an even number of reflections. Like Isom+ (R2 ), this is the “orientation-preserving” subgroup. But, unlike Isom+ (R2 ), Isom+ (S2 ) includes no “translations”—only rotations. We already know that the product of two reflections of S2 is a rotation. Hence, to show that the product of any even number of reflections is a rotation, it remains to show that the product of any two rotations of S2 is a rotation. Suppose that the two rotations of S2 are • a rotation through angle θ about point P (that is, a rotation with axis through P and its antipodal point −P), • a rotation through angle ϕ about point Q. We have established that a rotation through θ about P is the product of reflections in “lines” (great circles) through P. Moreover, they can be any “lines” L and M through P as long as the angle between L and M is θ /2. In particular, we can take the line M to go through P and Q. Similarly, a rotation through ϕ about Q is the product of reflections in any lines through Q meeting at angle ϕ /2, so we can take the first “line” to be M . The second “line” of reflection through Q is then the “line” N at angle ϕ /2 from M (Figure 7.5). If rL , rM , rN denote the reflections in L , M , N , respectively, then rotation through θ about P = r M rL , rotation through ϕ about Q = r N rM . (Bear in mind that products of transformations are read from right to left, as this is the order in which functions are applied.) Hence, the product of these rotations is rN rM rM rL = rN rL ,
because rM rM is the identity.
158
7 Transformations
L
P
θ /2
R
N
ϕ /2
M
Q
Figure 7.5: Reflection “lines” on the sphere And it is clear from Figure 7.5 that r N rL is a rotation (about the point R where L meets N ).
Some special rotations Before trying to obtain an overview of the rotation group of the sphere, it is helpful to look at the rotation group of the circle, which is analogous but considerably simpler. The circle can be viewed as the unit one-dimensional sphere S 1 in R2 , and its rotations are products of reflections in lines through O. This circumstance is what makes the rotation group of the circle similar to the rotation group of the sphere. What makes it a lot simpler is the fact that each rotation of S1 corresponds to a point of S1 , because each rotation of S1 is determined by the point to which it sends the specific point (1, 0). In other words, each rotation of the circle corresponds to an angle, namely the angle between the initial and final positions of any line through O. Also, rotations of S1 commute, because rotation through θ followed by rotation through ϕ results in rotation through θ + ϕ , which is also the result of rotation through ϕ followed by rotation through θ .
7.6 Representing space rotations by quaternions
159
In contrast to S1 , a rotation of S2 depends on three numbers: two angles that give the direction of its axis, and the amount of turn about this axis. Thus, the rotations of S2 cannot correspond to the points of S2 , although they do correspond to the points of an interesting three-dimensional space, as we shall see in Section 7.6. Rotations of S2 generally do not commute, as can be seen by combining a quarter turn z1/4 around the z-axis with a half-turn x1/2 around the x-axis. Supposing that the quarter turn is in the direction that takes (1, 0, 0) to (0, 1, 0), we have z1/4
x1/2
(1, 0, 0) −→ (0, 1, 0) −→ (0, −1, 0), whereas
x1/2
z1/4
(1, 0, 0) −→ (1, 0, 0) −→ (0, 1, 0).
Exercises In the Euclidean plane R2 , the product of a rotation about a point P and a rotation about a point Q is not necessarily a rotation. 7.5.1 Give an example of two rotations of R2 whose product is a translation. 7.5.2 By imitating the construction of rotations of S2 via reflections, explain how to decide whether the product of two rotations of R2 is a rotation and, if so, how to find its center and angle. The group of all isometries of R2 , unlike the group of rotations of R2 about O, is not commutative. 7.5.3 Find a rotation and reflection of R2 that do not commute.
7.6 Representing space rotations by quaternions The most elegant (and practical) way to describe rotations of R 3 or S2 is with the help of the quaternions, which were introduced in Section 6.6. Because they appeared there only in exercises, we now review their basic properties for the sake of completeness. A quaternion is a 2 × 2 matrix of the form q=
a + ib c + id −c + id a − ib
,
where a, b, c, d ∈ R and i2 = −1.
160
7 Transformations
We also write q in the form q = a1 + bi + cj + dk, where 1 0 i 0 0 1 0 i 1= , i= , j= , k= . 0 1 0 −i −1 0 i 0 The various products of i, j, and k are easily worked out by matrix multiplication, and one finds for example that ij = k = −ji and i2 = −1. Because q corresponds to the quadruple (a, b, c, d) of real numbers, we can view q as a point in R4 . If p is an arbitrary point in R4 then the map sending p → pq multiplies all distances in R4 by |q|, the distance of q from the origin. To see why, notice that det q = a2 + b2 + c2 + d 2 = |q|2 . Then it follows from the multiplicative property of determinants that |pq|2 = det(pq) = (det p)(det q) = |p|2 |q|2
and hence |pq| = |p||q|.
It follows that, for any points p1 , p2 ∈ R4 , |p1 q − p2 q| = |(p1 − p2 )q| = |p1 − p2 ||q|. Hence, the distance |p1 − p2 | between any two points is multiplied by the constant |q|. In particular, if |q| = 1, then the map p → pq is an isometry of R4 . The map p → qp (which is not necessarily the same as the map p → pq, because quaternion multiplication is not commutative) is likewise an isometry when |q| = 1. These maps are useful for studying rotations of R 4 but, more surprisingly, also for studying rotations of R3 .
Rotations of (i, j, k)-space If p is any quaternion in (i, j, k)-space, p = xi + yj + zk,
where x, y, z ∈ R,
and if q is any nonzero quaternion, then it turns out that qpq−1 also lies in (i, j, k)-space. Thus, if |q| = 1, then the map p → qpq−1 defines an isometry of R3 , because (i, j, k)-space is just the space of real triples (x, y, z) and hence a copy of R3 .
7.6 Representing space rotations by quaternions
161
Moreover, any quaternion with |q| = 1 can be written in the form q = cos
θ θ + (li + mj + nk) sin , 2 2
where l 2 + m2 + n2 = 1,
and the isometry p → qpq−1 is a rotation of (i, j, k)-space through angle θ about the axis through 0 and li + mj + nk. These facts can be confirmed by calculation, but we verify them only for the special case in which the axis of rotation is in the i direction, and for special points p that easily determine the nature of the isometry. Notice how the angles θ /2 in q and q−1 combine to produce angle of rotation θ . Example. The map p → qpq−1 , where q = cos θ2 + i sin θ2 . First we check that any point xi on the i-axis is fixed by this map. θ θ θ θ −1 xi cos − i sin qxiq = cos + i sin 2 2 2 2 θ θ θ θ xi cos + x1 sin because i2 = −1 = cos + i sin 2 2 2 2 θ θ θ θ 2θ 2θ + sin + 1 sin cos − sin cos = xi cos 2 2 2 2 2 2 = xi. Next we check that the point j is rotated through angle θ in the (j, k)plane, to the point j cos θ + k sin θ . θ θ θ θ qjq−1 = cos + i sin j cos − i sin 2 2 2 2 θ θ θ θ j cos + k sin because ji = −k = cos + i sin 2 2 2 2 θ θ θ θ + k 2 sin cos because ik = j, ij = k = j cos2 − sin2 2 2 2 2 = j cos θ + k sin θ . It can be similarly checked that qkq−1 = −k sin θ + j cos θ . Hence the isometry p → qpq−1 is a rotation of the (j, k)-plane through θ , because this is certainly what such a rotation does to the points 0, j, and k, and we know from Section 3.7 that any isometry of a plane is determined by what it does to three points not in a line.
162
7 Transformations
Thus, the isometry p → qpq−1 of (i, j, k)-space leaves the i-axis fixed and rotates the (j, k)-plane through angle θ , so it is a rotation through θ about the i-axis. It should be emphasized that if the quaternion q represents a certain rotation of R3 , then so does the opposite quaternion −q, because qpq−1 = (−q)p(−q)−1 . Thus, rotations of R3 actually correspond to pairs of quaternions ±q with |q| = 1. This has interesting consequences when we try to interpret the group of rotations of R3 as a geometric object in its own right (Section 7.8).
Exercises The representation of space rotations by quaternions is analogous to the representation of plane rotations by complex numbers, which was described in Section 4.7. As a warmup for the study of a finite group of space rotations in Section 7.7, we look here at some finite groups of plane rotations and the geometric objects they preserve. We take the plane to be C, the complex numbers. 7.6.1 Consider the square with vertices 1, i, −1, and −i. There is a group of four rotations of C that map the square onto itself. These rotations correspond to multiplying C by which four numbers? 7.6.2 The cyclic group Cn is the group of n rotations that maps a regular n-gon onto itself. These rotations correspond to multiplying C by which n complex numbers? The noncommutative multiplication of quaternions is a blessing when we want to use them to represent space rotations, because we know that products of space rotations do not generally commute. Nevertheless, one wonders whether there is a reasonable commutative “product” operation on any Rn , for any n ≥ 3. “Reasonable” here includes the property |uv| = |u||v| that holds for products on R and R2 (the real and complex numbers), and the field axioms from Section 6.5. In particular, there should be a multiplicative identity: a point 1 such that |1| = 1 and u1 = u for any point u. Moreover, because n ≥ 3, we can find points i and j, also of absolute value 1, such that 1, i, and j are in mutually perpendicular directions from O. Figure 7.6 shows these points, together with their negatives. √ 7.6.3 Show that |1 + i| = 2 = |1 − i|, and deduce from the assumptions about the product operation that 2 = |1 − i2 |, which means that the point 1 − i2 is at distance 2 from O. 7.6.4 Show also that |i2 | = 1, so the point 1 − i2 is at distance 1 from 1. Conclude from this and Exercise 7.6.3 that i2 = −1.
7.7 A finite group of space rotations
163
j −1
i O
−i 1 1−i
1+i
−j
Figure 7.6: Points in perpendicular directions from O 7.6.5 Show similarly that u2 = −1 for any point u whose direction from O is perpendicular to the direction of 1, and whose absolute value is 1. 7.6.6 Given that i and j are in perpendicular directions, show (multiplying the whole space by i) that so are i2 = −1 and ij, and hence so too are 1 and ij. 7.6.7 Thus, ij is one point u for which u2 = −1, by Exercise 7.6.5. Deduce that −1 = (ij)2 = (ij)(ij) = jiij by the commutative and associative laws and show that this leads to the contradiction −1 = 1. Therefore, when n ≥ 3, there is no product on Rn that satisfies all the field axioms.
7.7 A finite group of space rotations R3 is home to the regular polyhedra, the remarkable symmetric objects discussed in Section 1.6 and shown in Figure 1.19. The best known of them is the cube, and the simplest of them is the tetrahedron, which fits inside the cube as shown in Figure 7.7. Also shown in this figure are some rotations of the tetrahedron that are called symmetries because they preserve its appearance. If we choose a fixed position of the tetrahedron—a “tetrahedral hole” in space as it were— then these rotations bring the tetrahedron to positions where it once again
164
7 Transformations 1/2 turn 1/3 turn
Figure 7.7: The tetrahedron and axes of rotation fits in the hole. Altogether there are 12 such rotations. We can choose any one of the four faces to match a fixed face of the hole, say, the front face. Each of the four faces that can go in front has three edges that can match a given edge, say, the bottom edge, in the front face of the hole. Thus, we have 4 × 3 = 12 ways in which the tetrahedron can occupy the same position, each corresponding to a different symmetry. But once we have chosen a particular face to go in front, and a particular edge of that face to go on the bottom, we know where everything goes, so the symmetry is completely determined. Hence, there are exactly 12 rotational symmetries. Each symmetry can be obtained, from a given initial position, by rotations like those shown in Figure 7.7. First there is the trivial rotation, which gives the identity symmetry, obtained by rotation through angle zero (about any axis). Then there are 11 nontrivial rotations, divided into two different types: • The first type is a 1/2 turn about an axis through centers of opposite edges of the tetrahedron (which also goes through opposite face centers of the cube). There are three such axes. Hence, there are three rotations of this type. • The second type is a 1/3 turn about an axis through a vertex and the center of the face opposite to it (which also goes through opposite
7.7 A finite group of space rotations
165
vertices of the cube). There are four such axes, and hence eight rotations of this type—because the 1/3 turn clockwise is different from the 1/3 turn anticlockwise. Notice also that each 1/2 turn moves all four vertices, whereas each 1/3 turn leaves one vertex fixed and moves the remaining three. Thus, the 11 nontrivial rotations are all different. Therefore, together with the trivial rotation, they account for all 12 symmetries of the tetrahedron.
The quaternions representing rotations of the tetrahedron As explained in Section 7.6, a rotation of (i, j, k)-space through angle θ about axis li + mj + nk corresponds to a quaternion pair ±q, where q = cos
θ θ + (li + mj + nk) sin . 2 2
If we choose coordinate axes so that the sides of the cube in Figure 7.7 are parallel to the i, j, and k axes, then the axes of rotation are virtually immediate, and the corresponding quaternions are easy to work out. • We can take the lines through opposite face centers of the cube to be the i, j, and k axes. For a 1/2 turn, the angle θ = π , and hence θ /2 = π /2. Therefore, because cos π2 = 0 and sin π2 = 1, the 1/2 turns about the i, j, and k axes are given by the quaternions i, j, and k themselves. Thus, the three 1/2 turns are represented by the three quaternion pairs ±i,
±j,
±k.
• Given the choice of i, j, and k axes, the four rotation axes through opposite vertices of the cube correspond to four quaternion pairs, which together make up the eight combinations 1 √ (±i ± j ± k) (independent choices of + or − sign). 3 The factor √13 is to give each of these quaternions the absolute value 1, as specified for the representation of rotations.
166
7 Transformations For each 1/3 turn, we have θ = ±2π /3. Hence,
√ θ π 3 sin = ± sin = ± . 2 3 2 √ √ The 3 in sin π3 neatly cancels the factor 1/ 3 in the axis of rotation, and we find that the eight 1/3 turns are represented by the eight pairs of opposites among the 16 quaternions
θ π 1 cos = cos = , 2 3 2
1 i j k ± ± ± ± . 2 2 2 2 Finally, the identity rotation is represented by the pair ±1, and thus the 12 symmetries of the tetrahedron are represented by the 24 quaternions ±1,
±i,
±j,
±k,
1 i j k ± ± ± ± . 2 2 2 2
The 24-cell These 24 quaternions all lie at distance 1 from O in R4 , and they are distributed in a highly symmetrical manner. In fact, they are the vertices of a four-dimensional figure analogous to a regular polyhedron—called a regular polytope. This particular polytope is called the 24-cell. Because we cannot directly perceive four-dimensional figures, the best we can do is study the 24-cell via projections of it into R3 (just as we often study polyhedra, such as the tetrahedron and cube, via projections onto the plane such as Figure 7.7). One such projection is shown in Figure 7.8 (which of course is a projection of a three-dimensional figure onto the plane—but it is easy to visualize what the three-dimensional figure is). This superb drawing is taken from Hilbert and Cohn-Vossen’s Geometry and the Imagination.
Exercises The vertices of the 24-cell include the eight unit points (positive and negative) on the four axes in R4 , but the other 16 points have some less obvious properties. 7.7.1 Verify directly that the 16 points ± 12 ± 2i ± 2j ± k2 are all at distance 1 from the origin in R4 . 7.7.2 Deduce that the distance from the center to any vertex of a four-dimensional cube is equal to the length of its side. 7.7.3 Show also that each of the points ± 21 ± 2i ± 2j ± k2 is at distance 1 from the four nearest unit points on the axes.
7.8 The groups S3 and RP3
167
Figure 7.8: The 24-cell
7.8 The groups S3 and RP3 The rotations of the tetrahedron, which were discussed in Section 7.7, vividly show that a group of rotations is itself a geometric object. This statement is just as true of the group of all rotations of S2 . In fact, this group is closely related to two important geometric objects: the 3-sphere S3 and the three-dimensional real projective space RP3 . Just as the 1- and 2-spheres are the sets of points at unit distance from O in R2 and R3 respectively, the 3-sphere is the set of points in R4 at unit distance from O: S3 = {(a, b, c, d) ∈ R4 : a2 + b2 + c2 + d 2 = 1}. The points (a, b, c, d) on S3 correspond to quaternions q = a1+bi+cj+dk with |q| = 1, because |q|2 = a2 + b2 + c2 + d 2 . Hence, rotations of S2 , which correspond to pairs ±q of such quaternions, correspond to point pairs ±(a, b, c, d) on S3 .
168
7 Transformations
And to what else do the point pairs ±(a, b, c, d) correspond? Well, remember from Section 5.4 how we described the real projective space RP3 . Its “points” are lines through O in R4 . But a line through O in R4 meets S3 in a pair of antipodal points ±(a, b, c, d). Thus, it is also valid to view the point pairs ±(a, b, c, d) on S3 as single “points” of RP3 . Hence, rotations of S2 correspond to points of RP3 , and the group of all rotations of S2 is in some sense the “same” as the geometric object RP3 . To explain what we mean by “sameness” here, we have to say something about groups in general, what it means for two groups to be the “same”, and what it means for a geometric object to acquire the structure of a group.
Abstract groups and isomorphisms We began this chapter with the idea of a group of transformations: a collection G of functions on a space S with the properties that • if f , g ∈ G, then f g ∈ G, • if f ∈ G, then f −1 ∈ G. The “product” f g of f and g here is the composite function, which is defined by f g(x) = f (g(x)). However, we have found it convenient to represent certain functions, such as rotations, by algebraic objects, such as matrices, whose “product” is defined algebraically. It is therefore desirable to have a more general concept of group, which does not presuppose that the product operation is function composition. We define an abstract group to be a set G, which contains a special element 1 and for each g an element g−1 , with a “product” operation satisfying the following axioms: g1 (g2 g3 ) = (g1 g2 )g3 g1 = g gg
−1
=1
(associativity) (identity) (inverse)
The associative axiom is automatically satisfied for function composition, because if g1 , g2 , g3 are functions, then g1 (g2 g3 ) and (g1 g2 )g3 both mean the same thing, namely, the function g1 (g2 (g3 (x))). It is also satisfied when the group consists of numbers, because the product of numbers is well known to be associative.
7.8 The groups S3 and RP3
169
In other cases, the easiest way to prove associativity is to show, if possible, that the group operation corresponds to function composition. For example, the matrix product operation is associative because matrices behave like linear transformations under composition. It follows in turn that the quaternion product operation is associative, because quaternions can be viewed as matrices. When we say that the elements of a certain group G “correspond to” or “behave like” or “can be viewed as” elements of another group G , we have in mind a precise relationship called an isomorphism of G onto G . The word “isomorphism” comes from the Greek for “same form,” and it means that there is a one-to-one correspondence between G and G that preserves products. That is, an isomorphism is a function
ϕ : G → G
such that
ϕ (g1 g2 ) = ϕ (g1 )ϕ (g2 ).
For example, the group G of rotations of the circle, under composition of isometries, is isomorphic to group G of complex numbers of the form cos θ + i sin θ ,
under multiplication.
If rθ denotes the rotation through angle θ , then the isomorphism ϕ is defined by ϕ (rθ ) = cos θ + i sin θ . Sometimes there is a natural one-to-one correspondence ϕ between a group G and a set S. In that case, we can use ϕ to transfer the group structure from G to S. That is, we define the product of elements ϕ (g 1 ) and ϕ (g2 ) to be ϕ (g1 g2 ). Here are some examples. • The complex numbers cos θ + i sin θ form a group, and they correspond to the points (cos θ , sin θ ) of the unit circle S1 . Therefore, we can define the product of points (cos θ1 , sin θ1 ) and (cos θ2 , sin θ2 ) on S1 to be the point corresponding to the product of the corresponding complex numbers. This point is (cos(θ1 + θ2 ), sin(θ1 + θ2 )). • Likewise, the quaternions q = a1 + bi + cj + dk with |q| = 1 form a group, and they correspond to the points (a, b, c, d) of the 3-sphere S3 . Hence, we can define the product of points (a1 , b1 , c1 , d1 ) and (a2 , b2 , c2 , d2 ) corresponding to the quaternions q1 and q2 , say, to be the point corresponding to the quaternion q1 q2 . Under this product operation, S3 is a group.
170
7 Transformations • Finally, pairs of opposite quaternions ±q with |q| = 1 form a group under the operation defined by (±q1 )(±q2 ) = ±q1 q2 . We know that these pairs are in one-to-one correspondence with the points of RP3 . Hence, we can transfer the group structure of these quaternion pairs (which is also the group structure of the rotations of S2 ) to RP3 . Under the transferred operation, RP3 is a group.
The group structures on S1 , S3 , and RP3 obtained in this way are particularly interesting because they are continuous. That is, if g 1 is near to g1 and g2 is near to g2 , then g1 g2 is near to g1 g2 . It is known that S2 does not have a continuous group structure, and in fact S1 and S3 are the only spheres with continuous group structures on them.
7.9 Discussion The word “geometry” comes from the Greek for “earth measurement,” and legend has it that the subject grew from the land measurement concerns of farmers whose land was periodically flooded by the river Nile. As recently as the 18th century, one finds carpenters and other artisans listed among the subscribers to geometry books, so there is no doubt that Euclidean geometry is the geometry of down-to-earth measurement. It continues to be a tactile subject today, when one talks about “translating,” “rotating,” and “moving objects rigidly.” The most visual branch of geometry is projective geometry, because it is more concerned with how objects look than with what they actually are. It is no surprise that projective geometry originated from the concerns of artists, and that many of its practitioners today work in the fields of video games and computer graphics. Affine geometry occupies a position in the middle. It also originates from an artistic tradition, but from one less radical than that of Renaissance Italy—the classical art of China and Japan. Chinese and Japanese drawings often adopt unusual viewpoints, where one might expect perspective, but they generally preserve parallels. Typically, the picture is a “projection from infinity,” which is an affine map. Figure 7.9 shows an example, a woodblock print by the Japanese artist Suzuki Harunobu from around 1760.
7.9 Discussion
171
Figure 7.9: Harunobu’s Courtesan on veranda
Notice that all the parallel lines are shown as parallel, with the result that the (obviously rectangular) panels on the screen appear as identical parallelograms. Likewise, the planks on the veranda appear with parallel edges and equal widths, which creates a certain “flatness” because all parts of the picture seem to be the same distance away from us. Speaking mathematically, they are—because the view is what one would see from infinity with infinite magnification. A similar effect occurs in photographs of distant buildings taken with a large amount of zoom. Affine maps are also popular in engineering drawing, in which the socalled “axonometric projection” is often used to depict an object in three dimensions while retaining correct proportions in a given direction. The
172
7 Transformations
affine picture gives a good compromise between a realistic view and an accurate plan. See Figure 7.10, which shows an axonometric projection of a cube.
Figure 7.10: Affine view of the cube
The fourth dimension The discovery of quaternions in 1843 was the first of a series of discoveries that drew attention to spaces of more than three dimensions and to the remarkable properties of R4 in particular. From around 1830, the Irish mathematician William Rowan Hamilton had been searching in vain for “n-dimensional number systems” analogous to the real numbers R and the complex numbers C. Because C can be viewed as R2 under vector addition (u1 , u2 ) + (v1 , v2 ) = (u1 + v1 , u2 + v2 ) and the multiplication operation (u1 , u2 )(v1 , v2 ) = (u1 v1 − u2 v2 , u1 v2 + u2 v1 ), Hamilton thought that R3 could also be viewed as a number system by some clever choice of multiplication rule. He took a “number system” to be what we now call a field, together with an absolute value |u| = |(u1 , u2 , u3 )| = u21 + u22 + u23 which is multiplicative: |uv| = |u||v|.
7.9 Discussion
173
We now know (for example, by Exercise 7.6.7) that such a system is impossible in any Rn with n ≥ 3. But, luckily for Hamilton, it is almost possible in R4 . The quaternions satisfy all the field axioms except commutative multiplication, and their absolute value is multiplicative. The only other Rn that comes close is R8 , where the octonions O satisfy all the field properties except the commutative and associative laws. (Recall from Section 6.8 that the quaternions and octonions also play an important role in projective geometry.) Hamilton knew that quaternions give a nice representation of rotations in R3 , but the first to work out the quaternions for the symmetries of regular polyhedra was Cayley in 1863. Cayley’s enumeration of these quaternions may be found in his Mathematical Papers, volume 5, p. 529. The five regular polyhedra actually exhibit only three types of symmetry—because the cube and octahedron have the same symmetry type, as do the dodecahedron and the icosahedron—which therefore correspond to three highly symmetric sets of quadruples in R4 . Cayley did not investigate the geometric properties of these point sets in R4 , but in fact they were already known to the Swiss geometer Ludwig Schl¨afli in 1852. As we have seen, the 12 rotations of the tetrahedron correspond to the 24 vertices of a figure called the 24-cell. It gets this name because it is bounded by 24 identical regular octahedra. It is one of six regular figures in R4 , analogous to the regular polyhedra in R3 , called the regular polytopes. They were discovered by Schl¨afli, who also proved that there are regular figures analogous to the tetrahedron, cube, and octahedron in each Rn , but that R3 and R4 are the only Rn containing other regular figures. The 24-cell is the simplest of the exceptional regular figures in R 4 ; the other two are the 120-cell (bounded by 120 regular dodecahedra) and the 600-cell (bounded by 600 regular tetrahedra). The 120-cell has 600 vertices, which correspond to the cell centers of the 600-cell, and vice versa, so the two are related “dually” like the dodecahedron and the icosahedron. Moreover, the 600-cell arises from the icosahedron in the same way that the 24-cell arises from the tetrahedron. Its 120 vertices correspond to 60 pairs of opposite quaternions, each representing a rotational symmetry of the icosahedron. For more on these amazing objects, see my article The story of the 120-cell, which can be read online at http://www.ams.org/notices/200201/fea-stillwell.pdf
8 Non-Euclidean geometry P REVIEW In previous chapters, we have seen several reasons why there is such a subject as “foundations of geometry.” Geometry is fundamentally visual; yet it can be communicated by nonvisual means: by logic, linear algebra, or group theory, for example. The several ways to communicate geometry give several foundations. But also, there is more than one geometry. Section 7.4 gave a hint of this when we briefly discussed the geometry of the sphere in the language of “points” and “lines.” It seems reasonable to call great circles “lines” because they are the straightest curves on the sphere; but they certainly do not have all of the properties of Euclid’s lines. This characteristic makes geometry on a sphere a non-Euclidean geometry—one that has been known since ancient times. But it was never seen as a challenge to Euclid, probably because the geometry of the sphere is simply a part of three-dimensional Euclidean geometry, where great circles coexist with genuine straight lines. The real challenge to Euclid emerged from disquiet over the parallel axiom. Many people found it inelegant and wished that it was a consequence of Euclid’s other axioms. It is not, because there is a geometry that satisfies all of Euclid’s axioms except the parallel axiom. This is the geometry of a surface called the non-Euclidean plane. The non-Euclidean plane is not an artificial construct built only to show that the parallel axiom cannot be proved. It arises in many places, and today one can hardly discuss differential geometry, the theory of complex numbers, and projective geometry without it. In this chapter, we will see how it arises from the real projective line.
174
8.1 Extending the projective line to a plane
175
8.1 Extending the projective line to a plane In this book, we have been concerned mainly with the geometry of lines, partly because lines are the foundation of geometry and partly because lines are remarkably interesting. From the regular polygons to the Pappus and Desargues configurations, figures built from lines reveal beautiful connections between geometry and other parts of mathematics. We have seen some of these connections but have barely begun to explore them in depth. In fact we have not yet gone far toward understanding the geometry of even one line—the real projective line RP1 . In Chapter 5, we arrived at an algebraic summary of RP1 by representing its transformations as linear fractional functions f (x) =
ax + b , cx + d
where a, b, c, d ∈ R and ad − bc = 0,
(*)
and by uncovering the cross-ratio, a “ratio of ratios” left invariant by all linear fractional transformations. But this summary is not as geometric as one would like. It is hard (although perhaps not impossible) to “see” the cross-ratio, and indeed it is hard to see geometric phenomena on the line at all. If only we could extend the projective line in another dimension so that we could see it as a plane! Amazingly, this is possible, and the present chapter shows how. The idea is to let RP1 be the boundary (“at infinity”) of a plane whose transformations extend the linear fractional transformations of RP 1 in a natural way. Algebra suggests how this should be done. It suggests replacing the real variable x in the linear fractional transformations (*) by a complex variable z and interpreting f (z) =
az + b cz + d
as a transformation of the plane C of complex numbers. This idea needs a little modification. We should really use transformations of the upper half plane of complex numbers z = x + iy with y > 0, because the line of real numbers divides C into two halves. Either half can be taken as the “plane” bounded by the real line, but, when we want to transform one particular half, the extension from x to z is not always the obvious one. For example, the transformation x → −x of the line should not be extended to the transformation z → −z, because the latter maps the
176
8 Non-Euclidean geometry
upper half plane onto the lower. The correct extension is described in Section 8.2. Nevertheless, the extension from real to complex numbers works amazingly well. The extended transformations leave invariant a geometric quantity that is clearly visible, namely angle. The cross-ratio is also invariant (well, almost), for much the same algebraic reasons as before. And it gives us an invariant length in the half plane—something we certainly do not have in the projective line RP1 . The concept of length that emerges in this way is a little subtle, and it is not as easily visible as angle. It is a non-Euclidean measure of length, and it gives rise to non-Euclidean lines, which turn out to be the ordinary lines of the form x = constant and the semicircles with their centers on the x-axis. These “lines” are the curves of shortest non-Euclidean length between given points, and they have all the properties of “lines” in Euclid’s geometry except the parallel property. That is, if L is any non-Euclidean line and P is a point of the upper half plane outside L then there is more than one non-Euclidean line through P that does not meet L . Figure 8.1 shows an example. P
L
Figure 8.1: Failure of the parallel axiom for non-Euclidean “lines” The dotted line represents the real line y = 0 so the half plane y > 0 consists of the points strictly above it. The semicircle L is one “line” in the half plane, and the two semicircles passing through the point P clearly do not meet L , so they are two “parallels” of L . In the remainder of this chapter we explain in more detail why these semicircles should be regarded as “lines,” and why they satisfy all of Euclid’s axioms for lines except the parallel axiom.
8.1 Extending the projective line to a plane
177
Thus, the complex half plane not only allows us to visualize the geometry of the projective line; it also answers a fundamental question in the foundations of geometry by showing that the parallel axiom does not follow from Euclid’s other axioms.
The non-Euclidean “line” through two points One property of non-Euclidean “lines” can be established immediately. There is a unique non-Euclidean “line” through any two points (and hence non-Euclidean “lines” satisfy the first of Euclid’s axioms). • If the two points lie on the same vertical line x = l, then x = l is a nonEuclidean “line” containing them. And it is the only one, because a semicircle with its endpoints on the x-axis has at most one point on each line x = l. • If the two points P and Q do not lie on the same vertical line, there is a unique point R on the x-axis equidistant for both of them, namely, where the equidistant line of P and Q meets the x-axis. Then the semicircle with center R through P and Q is the unique non-Euclidean “line” through P and Q.
Exercises One can begin to understand the geometric significance of linear fractional transformations of the half plane by studying the simplest ones, z → z + l and z → kz for real k and l. 8.1.1 Show that the transformations z → z + l and z → kz (for k > 0) map the upper half plane onto itself and that they map “lines” to “lines.” 8.1.2 Explain why this is not the case when k and l are not real. 8.1.3 Show how to map the semicircle x2 + y2 = 1, y > 0, onto the semicircle (x − 1)2 + y2 = 4, y > 0, by a combination of transformations z → z + l and z → kz. 8.1.4 More generally, explain why any semicircle with center on the x-axis can be mapped onto any other by a combination of transformations z → z + l and z → kz. What is not yet clear is why semicircles should be regarded as “lines.” Their “linelike” behavior stems from the transformation z → 1/z, which we study in Section 8.2.
178
8 Non-Euclidean geometry
8.2 Complex conjugation We know from Section 5.6 that all linear fractional transformations of RP 1 are products of the transformations x → x + l, x → kx, and x → 1/x for real constants k = 0 and l. We called these the generating transformations of RP1 . The transformation x → x + l obviously extends to the transformation z → z + l, which maps the upper half plane onto itself for any real l, but the appropriate extension of x → kx is z → kz only for k > 0, because z → kz does not map the upper half plane onto itself when k < 0. In particular, what is the appropriate extension of x → −x to a map of the upper half plane? Geometrically, the answer is obvious. The transformation x → −x is reflection of the line in O, so its most appropriate extension is reflection of the half plane in the y-axis, that is, the transformation x + iy → −x + iy. This transformation can be expressed more simply with the help of the complex conjugate z, which is defined as follows. If z = x + iy, then z = x − iy. Then the reflection of z in the y-axis is −z because −z = −(x + iy) = −(x − iy) = −x + iy. Thus, the appropriate extension of x → −x is z → −z (Figure 8.2). y
−z
−x
z
O
x
Figure 8.2: Extending reflection from the line to the half plane More generally, the appropriate extension of x → kx when k < 0 is z → kz, the product of the reflection z → −z with the map z → |k|z (dilation by factor |k|).
8.2 Complex conjugation
179
A similar problem arises when we want to extend the transformation x → 1/x of RP1 to the half plane. The appropriate extension is not z → 1/z because this transformation does not map the upper half plane onto itself. In fact, if we write z in its polar form z = r(cos θ + i sin θ ), then 1 1 1 = (cos θ − i sin θ ) = (cos(−θ ) + i sin(−θ )) z r r because cos(−θ ) = cos θ and sin(−θ ) = − sin θ . Thus, z (at angle θ ) and 1/z (at angle −θ ) have opposite slopes from O. Hence, they lie in different half planes. The appropriate extension of x → 1/x is z → 1/z, which sends 1 1/z = (cos θ + i sin θ ) r lying in the same direction θ from O (Figure 8.3). This transformation is called reflection (or inversion) in the unit circle. z = r(cos θ + i sin θ ) to
y z
1/z
θ O
1/x 1
x
Figure 8.3: Extending inversion from the line to the half plane 1 Because all transformations x → ax+b cx+d of RP are products of x → x + l, x → kx, and x → 1/x, their extensions to the half plane are products of
• the horizontal translations z → z + l, • the dilations z → kz for k > 0, • reflection in the y-axis z → −z, • reflection in the unit circle z → 1/z. We call these the generating transformations of the half plane.
180
8 Non-Euclidean geometry
Equations of non-Euclidean “lines” Complex conjugation not only enables us to express reflection in lines and circles; it also enables us to write the equations of non-Euclidean “lines” very simply as equations in z. • First consider “lines” that are actual Euclidean lines, namely those of the form x = a, where a is a real number. An arbitrary point on this line is of the form z = a + iy, so z = a − iy, and z therefore satisfies the equation (*) z + z = 2a. • Next consider “lines” that are semicircles with centers on the x-axis. If the center is c and the radius is r, then any z on the circle satisfies |z − c| = r,
or equivalently, |z − c|2 = r2 .
But now notice that for any complex number x + iy we have |x + iy|2 = x2 + y2 = (x + iy)(x − iy) = (x + iy)(x + iy). Hence,
|z − c|2 = (z − c)(z − c) = (z − c)(z − c)
and the equation |z − c|2 = r2 becomes (z − c)(z − c) = r 2 , that is,
zz − cz − cz + cc = r 2 .
Finally, because c is a real number, we have c = c, so the equation can be written as zz − c(z + z) + c2 − r2 = 0.
(**)
The equations (*) and (**) are both of the form Azz + B(z + z) +C = 0 for some A, B,C ∈ R.
(***)
Conversely, if A and B are not both zero, then (***) reduces to one of the equations (*) or (**) above, if it is satisfied by any points z at all.
8.2 Complex conjugation
181
• If A = 0, then (***) becomes z + z + C/B = 0, which is (*) with 2a = −C/B. • If A = 0, then (***) becomes zz + (z + z)B/A + C/A = 0, which is (**) with c = −B/A and c2 − r2 = C/A if r2 = B2 /C2 − C/A ≥ 0. If r2 < 0, then no points z satisfy the equation (***), because the equation is equivalent to |z − c|2 = r2 and |z − c|2 is necessarily > 0. Thus, equations of non-Euclidean “lines” are the satisfiable equations Azz + B(z + z) +C = 0,
where
A, B,C ∈ R are not all zero.
Exercises I expect that most readers of this book are familiar with the complex numbers, but it still seems worthwhile to review the properties of the complex conjugate. Its role in geometric transformations may not be familiar, so we develop the basic facts from first principles. 8.2.1 Writing z1 as x1 + iy1 and z2 as x2 + iy2 , show that z1 + z2 = z1 + z2 and z1 z2 = z 1 z2 . 8.2.2 Similarly, show that 1/z = 1/z. 8.2.3 Deduce from Exercises 8.2.1 and 8.2.2 that, for any a, b, c, d ∈ R and z ∈ C, az+b the complex conjugate of az+b cz+d is cz+d . With these facts established, we are in a position to determine the extension to 1 the half plane of each linear fractional transformation x → ax+b cx+d of RP . What we know so far is that the extension of x → x + l is z → z + l, the extension of x → kx is z → kz when k > 0 and z → kz when k < 0, and that the extension of x → 1/x is z → 1/z. We also know that any transformation x → ax+b cx+d is a product of these generating transformations. Hence, the extension of x → ax+b cx+d to the half plane is the product of the corresponding extensions. It seems likely that the latter product az+b is either z → az+b cz+d or z → cz+d , so the main problem is to decide when the product az+b is z → cz+d and when it is z → az+b cz+d . 8.2.4 Write each generating transformation of RP1 in the form x → ax+b cx+d , and hence, show that those whose extension involves z are precisely those for which ad − bc < 0. 8.2.5 Deduce from Exercise 8.2.4 and Exercise 5.6.3 that the extension of a prodx+b1 x+b2 uct, of transformations x → ac11x+d and x → ac22x+d , is the product of their 1 2 extensions. 8.2.6 Deduce from Exercise 8.2.5, or otherwise, that the extension of x → az+b z → az+b cz+d when ad − bc > 0 and z → cz+d otherwise.
ax+b cx+d
is
182
8 Non-Euclidean geometry
It may seem unfortunate that the extension of x → ax+b cx+d is one of two different types: a function of z or a function of z. However, these two algebraic types are inevitable because they reflect a geometric distinction: The functions of z are orientation-preserving, and the functions of z are not. In particular, the linear fractional transformations z → az+b cz+d with ad − bc > 0 are precisely the orientation-preserving transformations of the half plane.
8.3 Reflections and M¨obius transformations 1 The extensions of the transformations x → ax+b cx+d from RP to the half plane could be called “linear fractional,” but this would be confusing, because one half of them are linear fractional functions of z and the other half are linear fractional functions of z. Instead they are called Mo¨ bius transformations, after the German mathematician August Ferdinand M o¨ bius. In 1855, M¨obius introduced a theory of transformations generated by reflections in circles, using the obvious generalization from reflection in the unit circle to reflection in an arbitrary circle. We will see below that all M o¨ bius transformations of the half plane are products of reflections. One advantage of the reflection idea is that it makes sense in three (or more) dimensions, where reflection in a sphere is meaningful but “linear fractional transformation” generally is not. It is also revealing to view the transformations of RP1 as the restrictions of M¨obius transformations of the half plane, as this brings to light a concept of “projective reflection”. Reflection in an arbitrary circle is defined by generalizing the relationship between z and 1/z shown in Figure 8.3. We say that points Q and Q are reflections of each other in the circle with center P and radius r if P, Q, Q lie in a straight line and |PQ||PQ | = r2 (Figure 8.4).
Q
Q P
r
Figure 8.4: Reflection in an arbitrary circle
8.3 Reflections and M¨obius transformations
183
If the circle (or, rather, its upper half) is a non-Euclidean line, then the center P lies on the x-axis, and reflection in this circle can be composed from generating transformations of the half plane as follows: • translate P to O, • reduce the radius to 1 by dilating by 1/r, • reflect in the unit circle, • restore the radius to r by dilating by r, • translate the center from O back to P. Likewise, reflection in an arbitrary vertical line, say x = a, can be composed from generating transformations of the half plane as follows: • translate the line x = a to the y-axis, • reflect in the y-axis, • translate the y-axis to the line x = a. Thus, all reflections in non-Euclidean lines are products of generating transformations of the half plane. Conversely, we now show that every generating transformation of the half plane is a product of reflections (and hence so is every transformation of the half plane). The generating transformations z → −z and z → 1/z are reflections by definition, so it remains to deal with the remaining generating transformations. • the horizontal translation z → z + l: this is a Euclidean translation, and it is the product of reflections in the lines x = 0 and x = l/2. • the dilation z → kz, where k > 0: this is the product of the reflection z → 1/z in the unit circle and the map √ z → k/z, which is reflection in the circle with center O and radius k. It should be mentioned that ordinary reflection—reflection in a straight line—is the limiting case of reflection in a circle obtained by letting P and r tend to infinity in such a way that the circle tends to a straight line. Because Euclidean lines are the fixed point sets of ordinary reflections, it is natural that the “lines” of the half plane should be the fixed point sets of its “reflections.”
184
8 Non-Euclidean geometry
Projective reflections Looking back from the half plane to its boundary line RP 1 , we realize that we now know more about projective transformations of the line than we did before. Any projective transformation of RP1 is a product of projective reflections, where a projective reflection is the restriction, to RP 1 , of a reflection of the half plane. There is a “three reflections theorem” for RP1 , analogous to the three reflections theorem for isometries of the Euclidean plane (Section 3.7). This follows from a three reflections theorem for the half plane, similar to the one for the Euclidean plane, that we will prove in Section 8.8.
Exercises The simplest reflections of RP1 are ordinary reflection in O, x → −x, and the restriction of reflection in the unit circle, x → 1/x. The map x → 1/x might be called “reflection in the point-pair {−1, 1},” and it generalizes to “reflection in the point-pair {a, b}.” (A point-pair {a, b} is a “0-dimensional sphere,” because it consists of the points at constant distance (b − a)/2 from the “center” (a + b)/2.) 8.3.1 Write down the formula for ordinary reflection in the point x = a. 8.3.2 Explain why the map x → c2 /x is reflection in the point-pair {−c, c}. 8.3.3 Using Exercise 8.3.2, or otherwise, show that reflection in the point-pair {a, b} is given by the linear fractional function f (x) =
x(a + b) − 2ab . 2x − (a + b)
8.3.4 Show that, as b → ∞, the function for reflection in the point-pair {a, b} tends to the function for ordinary reflection in the point x = a.
8.4 Preserving non-Euclidean lines 1 We have now extended the projective transformations x → ax+b cx+d of RP to az+b M¨obius transformations z → az+b cz+d or z → cz+d of the half plane, but are M¨obius transformations of the half plane any easier to understand? We intend to show that they are, by showing that they have more easily visible invariants than the transformations of RP1 . First we show the invariance of non-Euclidean lines, which we now define officially as the vertical lines x = constant and the semicircles with centers on the x-axis.
8.4 Preserving non-Euclidean lines
185
Each M¨obius transformation of the half plane maps non-Euclidean lines to non-Euclidean lines. For the generating transformations z → z + l, z → kz for k > 0, and z → −z, this is easy to see. Each of these transformations sends vertical lines to vertical lines, circles to circles, and the x-axis to the x-axis because: • z → z + l is a horizontal translation of the half plane. • z → kz with k > 0 is a dilation of the half plane by k. • z → −z is the Euclidean reflection of the half plane in the y-axis. Thus, any product of the three transformations just listed sends vertical lines to vertical lines and semicircles with centers on the x-axis to semicircles with centers on the x-axis. Hence, all products of the transformations z → z + l, z → kz for k > 0, and z → −z preserve non-Euclidean lines. To show that all M¨obius transformations preserve non-Euclidean lines, it therefore remains to show that reflection in the unit circle, z → 1/z, preserves non-Euclidean lines. This is less obvious, because reflection in a circle can send a vertical line to a semicircle and vice versa. We prove that non-Euclidean lines are preserved by using their equations (***) derived in Section 8.3. Given a non-Euclidean line, whose points z satisfy an equation Azz + B(z + z) +C = 0 for some A, B,C ∈ R,
(***)
we wish to find the equation satisfied by the points of its reflection in the unit circle. These are the points w = 1/z, so we seek the equation satisfied by w. The required equation is likely to involve w = 1/z as well, so we are looking for an equation connecting 1/z and 1/z. Such an equation is easy to find: just divide the equation (***) by zz. Division yields the equation C 1 1 + + = 0, A+B z z zz that is, Cww + B(w + w) + A = 0.
(****)
Equation (****) is satisfied by the reflections w = 1/z of the points z satisfying (***), and (****) has the same form as (***), because A, B,C ∈ R. Hence, (****) also represents a non-Euclidean line.
186
8 Non-Euclidean geometry
Exercises An example in which reflection in the unit circle sends a vertical line to a semicircle is shown in Figure 8.5.
1
O
Figure 8.5: Reflection of the line x = 1 8.4.1 Give intuitive reasons why the reflection of the line x = 1 in the unit circle should have one end at 1 on the x-axis and the other end at O. 8.4.2 Show that the line x = 1 has equation z + z = 2, and that its reflection in the unit circle has equation w + w = 2ww. 8.4.3 Verify that w + w = 2ww is the equation of the semicircle with ends O and 1 on the x-axis.
8.5 Preserving angle Next to non-Euclidean lines, the most visible invariant of M o¨ bius transformations is angle. Because non-Euclidean lines are not necessarily straight, the angle between two of them is really the angle between their tangents at the point of intersection. Nevertheless, it is easy to see the angle between non-Euclidean lines. Figure 8.6 shows an example. L
M
C 0
1
Figure 8.6: Some non-Euclidean lines and the angles between them
8.5 Preserving angle
187
The three non-Euclidean lines are the unit circle C , the vertical L where x = 1/2, and its reflection M in the unit circle, which happens to be the semicircle with endpoints 0 and 2 on the x-axis. At the point where the three non-Euclidean lines meet, they divide the space around the point into six equal angles, so each angle is 2π /6 = π /3. This equality is confirmed by the tangents, which are shown as dashed lines. Notice that any two of C , L , and M are reflections of each other in the third non-Euclidean line, so the figure shows numerous instances of an angle equal to its reflection. To show that any angle is preserved by any M¨obius transformation, we look once again at the properties of the generating transformations.
The effect of M¨obius transformations The M¨obius transformations z → z + l and z → −z are Euclidean isometries; hence, they certainly preserve angle (along with length, area, and so on). The M¨obius transformations z → kz for k > 0 are dilations; hence, they too preserve angle. Thus, it suffices to prove that angle is preserved by the remaining generator of Mo¨ bius transformations: reflection in the unit circle, z → 1/z. The latter transformation is the composite of z → −z and z → −1/z, so it suffices in turn to prove that z → −1/z preserves angle. We therefore concentrate our attention on the Mo¨ bius transformation z → −1/z. This transformation does not in general preserve Euclidean lines, because it may map them to circles. Thus, we need to be aware that “angle” generally means the angle between curves and hence the angle between the tangents. However, we can avoid computing the position of tangents by taking the infinitesimal view of angle. That is, we study what becomes of the direction between two points, z and z + ∆z, when we send them to −1/z and −1/(z + ∆z), respectively, and let ∆z tend to zero. If ∆z is the point at distance ε from O in direction θ , then ∆z = ε (cos θ + i sin θ ), because cos θ + i sin θ is the point at distance 1 from O in direction θ . It follows that the point at distance ε from z in direction θ is z + ∆z = z + ε (cos θ + i sin θ ), and that the point z + ∆z tends to z in the constant direction θ as ε tends to zero.
188
8 Non-Euclidean geometry
The difference between the image points −1/(z + ∆z) and −1/z is therefore 1 1 z + ε (cos θ + i sin θ ) − z − = z z + ε (cos θ + i sin θ ) z(z + ε (cos θ + i sin θ )) ε (cos θ + i sin θ ) . = z(z + ε (cos θ + i sin θ )) Now as ε tends to zero, this difference is ever more closely approximated by ε (cos θ + i sin θ ) . z2 To be precise, the direction from −1/z to −1/(z+∆z) tends to the direction of ε (cos θ + i sin θ )z−2 , which is θ +constant. The constant is the argument (angle) of z−2 , recalling from Section 4.7 that the argument of a product of complex numbers is the sum of their arguments. The angle between two smooth curves meeting at z (approximated by the difference in directions from z to points z + ∆1 z and z + ∆2 z on the respective curves) is therefore the angle between the images of these curves under the map z → −1/z. This is because a smooth curve is one for which the direction from z to z + ∆z tends to a constant θ as z + ∆z tends to z along the curve. Non-Euclidean lines are smooth, so the angle between them is preserved by the transformation z → −1/z, as required.
Tilings of the half plane If one takes a triangle with angles π /p, π /q, π /r, for some natural numbers p, q, r, then any reflection of that triangle will have angles π /p, π /q, π /r. Reflecting the reflections causes the space around each vertex to be exactly filled with corners of triangles. For example, the space around the vertex of angle π /p becomes filled with 2p corners of angle π /p. In fact, the whole half plane becomes filled, or tiled, by copies of the original triangle. An example is shown in Figure 8.7, where the basic tile has angles π /2, π /3, and π /7. Notice that the angle sum π /2 + π /3 + π /7 is less than π . In fact, the angle sum of any triangle bounded by non-Euclidean lines is less than π , and the quantity (π − angle sum) is proportional to the area of the triangle. This elegant result is less surprising when one learns that the area of spherical triangle is also proportional to π − angle sum (see exercises below).
8.5 Preserving angle
189
Figure 8.7: Tiling by repeated reflections However, it does reveal a limitation in the half plane view of non-Euclidean geometry: All the triangles in Figure 8.7 have equal non-Euclidean area, but they certainly do not look equal! One should think of the half plane as a kind of “perspective view” of the non-Euclidean plane with the x-axis as a horizon. The x-axis is infinitely distant, because there are infinitely many identical triangles between any point of the half plane and the x-axis. In this respect, the half plane is like a perspective view of a Euclidean tiled floor, except that ordinary perspective preserves straightness and distorts angle, whereas this “non-Euclidean perspective” distorts straightness and preserves angle. There are other views of the non-Euclidean plane that make non-Euclidean lines look straight (see Section 8.9), but any such view has a curved horizon! Another way in which a tiling of the half plane resembles a perspective view is that one can estimate the length of a line by counting the numbers of tiles that lie along it. There is indeed a non-Euclidean measure of distance that is invariant under M¨obius transformations, and we will see exactly what it is in Section 8.6.
190
8 Non-Euclidean geometry
Exercises The English mathematician Thomas Harriot discovered that the area of a spherical triangle is proportional to (angle sum − π ) in 1603. His argument is based on the two views of a spherical triangle shown in Figure 8.8.
C B
A α
B
β
A
∆αβ γ γ C
(a)
α
∆αβ γ
∆α (b) Figure 8.8: Area of a spherical triangle
8.6 Non-Euclidean distance
191
View (a) shows all sides of the spherical triangle extended to great circles. These divide the sphere into eight spherical triangles, which are obviously congruent in antipodal pairs. View (b) shows the result of extending two sides, which is a “slice” of the sphere with area proportional to the angle at its two ends. 8.5.1 Letting the area of the triangle with angles α , β , γ be ∆αβ γ , and letting the areas of the other triangles be ∆α , ∆β , ∆γ as shown in view (a), prove that 2 ∆αβ γ + ∆α + ∆β + ∆γ = area of sphere, call it A.
(1)
8.5.2 Use view (b) to explain why ∆αβ γ + ∆α =
α A, 2π
∆αβ γ + ∆β =
β A, 2π
∆αβ γ + ∆γ =
γ A. 2π
8.5.3 Deduce from Exercise 8.5.2 that 3∆αβ γ + ∆α + ∆β + ∆γ =
α +β +γ A 2π
8.5.4 Deduce from equations (1) and (2) that 4∆αβ γ =
(2)
α +β +γ −π A, and hence that π
∆αβ γ = constant × (α + β + γ − π ). 8.5.5 Using a formula for the area of the sphere, show that ∆αβ γ = α + β + γ − π on a sphere of radius 1.
8.6 Non-Euclidean distance So far we have found invariants of Mo¨ bius transformations by geometrically inspired guesses that can be confirmed by calculations with linear fractional functions. But still up our sleeve is the cross-ratio card, which carries the fundamental invariant of linear fractional transformations, and to find out what non-Euclidean distance is we finally have to play it. We know from Section 5.7 that the cross-ratio is invariant under the transformations x → x + l and x → kx, and exactly the same calculations apply to z → z + l and z → kz. It is invariant under z → −1/z, as can be shown by a calculation similar to, but shorter than, that given in Section 5.7 for x → 1/x. However, it is not generally invariant under the M o¨ bius transformation z → −z, because this replaces the cross-ratio by its complex conjugate. We can only say that Mo¨ bius transformations either leave the cross-ratio invariant or change it to its complex conjugate.
192
8 Non-Euclidean geometry
Luckily, this does not matter, because we are interested in the crossratio only when the four points lie on a non-Euclidean line. It turns out that the cross-ratio of four points on a non-Euclidean line is real, and hence equal to its own complex conjugate. This is obvious when the points are pi, qi, ri, si on the upper y-axis, because, in this case, the cross-ratio equals the real number (r−p)(s−q) (r−q)(s−p) by cancellation of the i factors. It follows for any other non-Euclidean line L by mapping the upper y-axis onto L by a Mo¨ bius transformation. • If L is another vertical line x = l, we map the upper y-axis to L by z → z + l. • If L is a semicircle with center on the x-axis, we first map the upper y-axis to x = 1 by z → z + 1, and then to the semicircle with ends at 0 and 1 by z → 1/z. Finally we map this semicircle to L by dilating it to the radius of L and then translating its center to the center of L. We know from the previous section that the transformations z → z + l, z → kz for k > 0, z → −z, and z → −1/z generate all Mo¨ bius transformations, so we have now proved that the cross-ratio of any four points on a non-Euclidean line is preserved by Mo¨ bius transformations. So far, so good, but distance is a function of two points, not four. If the cross-ratio is going to help us define distance, we need to specialize it to a function of two variables. One of the beauties of a non-Euclidean line is that it lies between two endpoints. The non-Euclidean line represented by the upper y-axis, for example, consists of the points between 0 and ∞. The endpoints are not points of the line, but it is meaningful to include them in a cross-ratio, because M¨obius transformations apply to all complex numbers, and ∞. If we take 0 and ∞ as the third and fourth members of the quadruple pi, qi, ri, si on the upper y-axis, then the cross-ratio of this quadruple simplifies as follows: (r − p)(s − q) (r − p)(1 − q/s) = dividing top and bottom by s (r − q)(s − p) (r − q)(1 − p/s) r− p = because s = ∞ and 1/∞ = 0 r−q p because r = 0. = q
8.6 Non-Euclidean distance
193
Any M¨obius transformation of the upper y-axis sends endpoints to endpoints, as one can see from the generating transformations, but it is possible for 0 and ∞ to be exchanged. If r = ∞ and s = 0, we find that the crossratio of pi, qi, ri, si is q/p, not p/q. Thus, q/p is not an invariant of M o¨ bius transformations of the upper y-axis, but | log qp | is, because log qp = − log qp for any p, q > 0. This prompts us to make the following definition. Distance on the upper y-axis. The non-Euclidean distance ndist(pi, qi) between points pi and qi on the upper y-axis is | log qp |. This definition of distance is appropriate for two reasons: • As already shown, non-Euclidean distance on the upper y-axis is invariant under all M¨obius transformations. • Non-Euclidean distance is additive. That is, if pi, qi, ri lie on the upper y-axis in that order, then ndist(pi, ri) = ndist(pi, qi) + ndist(qi, ri). This is because log r = log q r = log q + log r = log q + log r p p q p q p q by the additive property of the logarithm function. It follows from this definition that the infinity of points 2n i, for integers n, are equally spaced along the upper y-axis, in the sense of non-Euclidean distance. The faces shown in Figure 8.9 are of equal size in this sense. The upper y-axis is not only infinite in the upward direction, but also in the downward direction. There is infinite non-Euclidean distance between any of its points and the x-axis. Thus, the upper y-axis satisfies Euclid’s second axiom for “lines”: Any segment of it can be “extended indefinitely.” Having defined non-Euclidean distance on the upper y-axis, we can use the axis as a “ruler” to measure the distance between two points in the upper half plane. Given any two points u and v, we find the unique nonEuclidean line L through u and v as described in Section 8.1, and then map L onto the upper y-axis by a Mo¨ bius transformation f as described (in reverse) in the first part of this section. We take the non-Euclidean distance from u to v to be the non-Euclidean distance from f (u) to f (v), namely ndist( f (u), f (v)).
194
8 Non-Euclidean geometry
Figure 8.9: Faces of equal non-Euclidean size The quantity ndist( f (u), f (v)) does not depend on the Mo¨ bius transformation f used to map L onto the upper y-axis. If g is another M o¨ bius transformation mapping L onto the upper y-axis, then f g −1 is a M¨obius transformation that maps the upper y-axis onto itself and sends the points g(u) and g(v) to f (u) and f (v), respectively. Hence, ndist(g(u), g(v)) = ndist( f (u), f (v)) by the invariance of non-Euclidean distance on the upper y-axis under M¨obius transformations.
The hidden geometry of the projective line As we mentioned in Section 7.1, Klein associated a “geometry” with each group of transformations. We have set up the group of transformations of the half plane to be isomorphic to the group of transformations of RP 1 . Hence, the half plane and RP1 have isomorphic geometries in the sense of Klein, even though they seem very different. Indeed, we transferred geometry from RP1 to the half plane mainly because of the difference: Geometry is much more visible in the half plane. Figure 8.7 is one illustration of this, and Figure 8.10 is another—a regular tiling of the half plane by fish that are congruent in the sense of non-Euclidean length. Figure 8.10 is essentially the picture Circle Limit I, by M. C. Escher, but mapped to the half plane by the transformation z →
1 − zi . z−i
8.6 Non-Euclidean distance
195
Figure 8.10: Half plane version of Escher’s Circle Limit I By restricting M¨obius transformations to the boundary of the half plane, half plane geometry can be compressed into the geometry of RP 1 , even though RP1 has no concepts of length or angle. Conversely, length and angle emerge when RP1 is expanded to the half plane.
Exercises We can now confirm the impression given by Figure 8.7, that each non-Euclidean line is infinite in both directions, as demanded by Euclid’s second axiom. 8.6.1 Show that the y-axis, and hence any non-Euclidean line, can be divided into infinitely many segments of equal non-Euclidean length. 8.6.2 Find a M¨obius transformation sending 0, ∞ to −1, 1, respectively, and hence mapping the y-axis onto the unit semicircle. 8.6.3 Using the transformation found in Exercise 8.6.2, find an infinite sequence of points on the unit semicircle that are equally spaced in the sense of nonEuclidean length.
196
8 Non-Euclidean geometry
Supposing that the equal faces shown in Figure 8.9 have non-Euclidean width ε , which can be as small as we please, we can draw some interesting conclusions about the non-Euclidean distance between non-Euclidean lines. 8.6.4 Show that the non-Euclidean distance between the lines x = 0 and x = 1 tends to zero as y tends to ∞. 8.6.5 Show that the M¨obius transformation z → 2/(1 − z) sends the unit circle and the line x = 1 to the lines x = 1 and x = 0, respectively. 8.6.6 Deduce from Exercise 8.6.5 that the non-Euclidean distance between the unit circle and the line x = 1 tends to zero as these non-Euclidean lines approach the x-axis.
8.7 Non-Euclidean translations and rotations Like the Euclidean plane, the half plane has isometries called translations and rotations, which are products of two reflections. Their nature depends on whether the lines of reflection meet or have a common end. A translation is the product of reflections in non-Euclidean lines that do not meet and do not have a common end. A simple example is z → 2z, which √ is the product of reflections in the circles with center 0 and radii 1 and 2. This translation maps each face in Figure 8.9 to the one above it. Any non-Euclidean translation maps a unique non-Euclidean line, called the translation axis, into itself. Also mapped into themselves are the curves at constant non-Euclidean distance from the translation axis, which (for distance > 0) are not non-Euclidean lines. For z → 2z, the translation axis is the y-axis and the equidistant curves are the Euclidean lines y = ax. Each non-Euclidean line perpendicular to the translation axis is mapped onto another such line. Figure 8.11 shows the translation axis, two equidistant curves (in gray), and some of their perpendiculars (on the left when the axis is vertical, and on the right when it is not). Notice that the equidistant curves in general are Euclidean circles passing through the two ends of the translation axis. The translation moves each non-Euclidean perpendicular to the next. The product of reflections in two non-Euclidean lines that meet at a point P is a non-Euclidean rotation about P. The point P remains fixed and points at non-Euclidean distance r from P remain at non-Euclidean distance r from P, since reflection is a non-Euclidean isometry. Hence, these points move on a non-Euclidean circle of radius r. It turns out that a non-Euclidean circle is a Euclidean circle, although its non-Euclidean
8.7 Non-Euclidean translations and rotations
197
y
0 Figure 8.11: Non-Euclidean translations center (the point at constant non-Euclidean distance from all its points) is not its Euclidean center. For example, if we take the product of the reflection z → −z in the y-axis with the reflection z → −1/z in the unit circle, the result is a rotation through angle π about the point i where these two non-Euclidean lines meet. More generally, if we have two non-Euclidean lines through P meeting at angle θ , then the product of reflections in these lines is a rotation about P through angle 2θ . Figure 8.12 shows four non-Euclidean lines through i and two non-Euclidean circles (in gray) with non-Euclidean center at i. A rotation of π /4 about i moves each non-Euclidean line to the next and maps each circle into itself.
i
0 Figure 8.12: A non-Euclidean rotation about i
198
8 Non-Euclidean geometry
A limiting case of rotation is where the two lines of reflection do not meet in the half plane, but have a common end P on the boundary R ∪ {∞} at infinity. Here P is a fixed point, each non-Euclidean line ending at P is moved to another line ending at P, and each curve perpendicular to all these lines is mapped onto itself. This kind of isometry is called a limit rotation, and each curve mapped onto itself is called a limit circle or horocycle. The simplest example is the Euclidean horizontal translation z → z + 1, which is the product of reflections in the vertical lines x = 0 and x = 1/2. Each vertical line x = a is mapped to the line x = a + 1, and each horizontal line y = b is mapped onto itself. Thus, the horizontal lines y = b, which we know are not non-Euclidean lines, are limit circles.
0
0
1 Figure 8.13: Limit rotations
Like equidistant curves, limit circles can be Euclidean lines, but generally they are Euclidean circles. Figure 8.13 shows the exceptional case z → z + 1, where the limit circles are the Euclidean horizontal lines (in gray), and the typical case z → z/(1 − z), where the limit circles are the gray circles tangential to the boundary at the fixed point z = 0. As in the previous pictures, the isometry moves each non-Euclidean line to the next, and maps each gray curve onto itself.
Exercises 8.7.1 Check that the product of reflections in the y-axis and the unit circle is z → −1/z, and that i is the fixed point of this map. √ 8.7.2 Show also that z → −1/z maps each circle of the form |z − ti| = t 2 − 1 onto itself. The limit rotation z → z/(1 − z) above is obtained by moving the limit rotation z → z + 1 about ∞ to a limit rotation about 0 with the help of the rotation z → −1/z that exchanges 0 and ∞.
8.8 Three reflections or two involutions
199
8.7.3 If f (z) = z + 1 and g(z) = −1/z, show that g f g−1 (z) = z/(1 − z). 8.7.4 Describe in words what g−1 , f , g in succession do to the half plane, and hence explain geometrically why g f g−1 has fixed point 0.
8.8 Three reflections or two involutions It is possible to prove that each isometry of the half plane is the product of three reflections, following much the same approach as was used in Section 3.7 to prove the three reflections theorem for the Euclidean plane. The details of this approach are worked out in my book Geometry of Surfaces. However, our approach to isometries of the Euclidean plane began with a definition of Euclidean distance; we then had to find the transformations that leave it invariant. Here we know the isometries of the half plane— the M¨obius transformations—so the only problem is to express them as products in some simple way. To do this, we can interpret Mo¨ bius transformations on RP1 , and exploit known theorems of projective geometry. Surprisingly, there is a theorem about RP1 that goes one better than the three reflections theorem, namely the two involutions theorem from Veblen and Young’s 1910 book Projective Geometry, p. 223. An involution is a transformation f such that f 2 is the identity. Thus, the involutions include the reflections, but some other transformations as well, such as the function x → −1/x, which (when extended to the half plane) represents a half turn about the point i. The name “involution” is one of many terms introduced into projective geometry by Desargues, and it is the only one that has stuck. To pave the way for the two involutions theorem (and the three reflections theorem that follows from it), we first note three consequences of the results in Section 5.8 about transformations of RP1 . • Any four points p, q, r, s ∈ RP1 can be mapped to q, p, s, r, respectively, by a linear fractional transformation. Notice that [p, q : r, s] = [q, p : s, r] because (r − p)(s − q) (s − q)(r − p) = . (r − q)(s − p) (s − p)(r − q) Hence, by the “criterion for four-point maps” in Section 5.8, there is a linear fractional f mapping p, q, r, s to q, p, s, r, respectively.
200
8 Non-Euclidean geometry • If g is a linear fractional transformation that exchanges two points, then g is an involution. Suppose that p and q are two points with g(p) = q and g(q) = p. Let r be another point, not fixed by g, and suppose that g(r) = s. Because any linear fractional function is one-to-one, it follows that p, q, r, s are different. Hence, by the previous result, there is a linear fractional f mapping p, q, r, s to q, p, s, r, respectively. Because f agrees with g on the three points p, q, r, the functions f and g are identical by the “uniqueness of three-point maps” in Section 5.8. For any nonfixed point r of g, we therefore have g2 (r) = g(s) = f (s) = r, and if r is a fixed point, then g2 (r) = r obviously. Hence, g2 (x) = x for any x ∈ RP1 , and so g is an involution. • For any three points p, q, r, there is an involution that exchanges p, q and fixes r. By “existence of three-point maps” from Section 5.8, there is a linear fractional function g that sends p, q, r to q, p, r, respectively. Thus, g fixes r, and because it exchanges p and q, it is an involution by the previous result.
Two involutions theorem. Any linear fractional transformation h of RP 1 is the product of two involutions. If h = identity, then h = identity · identity, which is the product of two involutions. If not, let p be a point not fixed by h, so h(p) = r = p, and let h(r) = q. Then q = r, because h−1 is also a linear fractional transformation and hence one-to-one. If q = p, then h exchanges p and r. Hence, h is itself an involution by the second result above. We can therefore assume that p, q, r are three different points; in which case, the third result above gives a linear fractional involution f such that f (p) = q,
f (q) = p,
f (r) = r.
Also, f h exchanges the two points p and r because f h(p) = f (r) = r,
f h(r) = f (q) = p.
8.8 Three reflections or two involutions
201
Thus, f h is an involution too; call it g. Finally, applying f −1 to both sides of f h = g, we get h = f −1 g. So h is the product of two involutions, f −1 = f and g, as required.
We now consider M¨obius transformations of the half plane, each of which is the unique extension of a linear fractional transformation of RP 1 . Such a function is determined by its values at three points on RP 1 , by “uniqueness of three-point maps.” We use the same letter for a linear fractional transformation of RP1 and its extension to a M¨obius transformation of the half plane, and we systematically use the fact that Mo¨ bius transformations preserve non-Euclidean lines and angles. Three reflections theorem. Any M¨obius transformation of the half plane is the product of at most three reflections. The involution f in the proof above, which exchanges p, q and fixes r, necessarily maps the non-Euclidean line L from p to q into itself. Points of L near the end p are sent to points near the end q, and vice versa. It follows by continuity that some point u on L is fixed by f , and hence the unique non-Euclidean line M through u and ending at r is mapped into itself by f . Also, because any M¨obius transformation preserves angles, M must be perpendicular to L (Figure 8.14). Thus, f has the same effect on p, q, r as reflection in the line M , so f is this reflection by “uniqueness of three-point maps.” L
u
p
q Figure 8.14: Lines involved in the involution f
M
r
202
8 Non-Euclidean geometry
Now consider the involution g, which is associated with a similar pair of lines L and M . Only the names of their ends are different, but the reader is invited to draw them to keep track. By the argument just used for f , the involution g is a reflection if it has a fixed point on RP1 . In any case, g maps the line L with ends p and r into itself, exchanging the ends, so g has a fixed point u on L by the argument just used for f . Also, because g preserves angles, g maps the non-Euclidean line M through u and perpendicular to L into itself. Thus, if g has no fixed point on RP1 , it necessarily exchanges the ends s and t of M . But then g has the same effect on the three points p, r, s as the product of reflections in L and M , so g is this product of reflections, by “uniqueness of three-point maps” again. Thus, f g, which is an arbitrary M¨obius transformation by the theorem above, is the product of at most three reflections.
Exercises The argument above appeals to “continuity” to show the existence of a fixed point on a non-Euclidean line whose ends are exchanged by an involution. This argument is valid, and it may be justified by the intermediate value theorem, well known from real analysis courses. However, some readers may prefer an actual computation of the fixed point. One way to do it is as follows. Suppose f (x) = ax+b cx+d and that f (p) = q, f (q) = p. 8.8.1 Deduce that a = −d and b = cpq − a(p + q), so that f has the form f (x) =
a(x − p − q) + cpq k(x − p − q) + pq = cx − a x−k
8.8.2 Solve the equation x=
if c = 0.
k(x − p − q) + pq , x−k
and hence show that the fixed points of f are u = k ± (k − p)(k − q). 8.8.3 Assuming that (k − p)(k − q) < 0, so one fixed point is in the upper half plane, show that its distance from the center (p + q)/2 of the semicircle with ends p and q is |(p − q)/2|. 8.8.4 Deduce from Exercises 8.8.1–8.8.3 that f has a fixed point on the nonEuclidean line with ends p and q.
8.9 Discussion
203
8.9 Discussion The non-Euclidean parallel hypothesis It has often been said that the germ of non-Euclidean geometry is in Euclid’s own work, because Euclid recognized the exceptional character of the parallel axiom and used it only when it was unavoidable. Later geometers noted several plausible equivalents of the parallel axiom, such as • the equidistant curve of a line is a line, • the angle sum of a triangle is π , • similar figures of different sizes exist, but no outright proof of it from Euclid’s other axioms was found. On the contrary, attempts to derive a contradiction from the existence of many parallels—what we will call the non-Euclidean parallel hypothesis—led to a rich and apparently coherent geometry. This is the geometry we have been exploring in the half plane, now called hyperbolic geometry. Hyperbolic geometry diverges from Euclidean geometry in the opposite direction from spherical geometry—for example, the angle sum of a triangle is < π , not > π —but the divergence is less extreme. The “lines” of spherical geometry violate all three of Euclid’s axioms about lines, whereas the “lines” of hyperbolic geometry violate only the parallel axiom. The first theorems of hyperbolic geometry were derived by the Italian Jesuit Girolamo Saccheri in an attempt to prove the parallel axiom. In his 1733 book, Euclides ab omni naevo vindicatus (Euclid cleared of every flaw), Saccheri assumed the non-Euclidean parallel hypothesis, and sought a contradiction. What he found were asymptotic lines: lines that do not meet but approach each other arbitrarily closely. This discovery was curious, and more curious at infinity, where Saccheri claimed that the asymptotic lines would meet and have a common perpendicular. Finding this “repugnant to the nature of a straight line,” he declared a victory for Euclid. But the common perpendicular at infinity is not a contradiction, and indeed (as we now know) it clearly holds in the half plane. There are non-Euclidean lines that approach each other arbitrarily closely in nonEuclidean distance, such as the unit semicircle and the line x = 1, and they have a common perpendicular at infinity—the x-axis. Saccheri had unwittingly discovered not a bug, but a key feature of hyperbolic geometry.
204
8 Non-Euclidean geometry
The non-Euclidean geometry of the hyperbolic plane began to take shape in the early 19th century. A small circle of mathematicians around Carl Friedrich Gauss (1777–1855) explored the consequences of the nonEuclidean parallel hypothesis, although Gauss did not publish on the subject through fear of ridicule. Gauss was the greatest mathematician of his time, but he was unwilling to publish “unripe” work, and he evidently felt that non-Euclidean geometry lacked a solid foundation. He knew of no concrete interpretation, or model, of non-Euclidean geometry, and in fact, none was discovered in his lifetime. It is a great irony that some of his own discoveries—in the geometry of curved surfaces and the geometry of complex numbers—can provide such models. The first to publish comprehensive accounts of non-Euclidean geometry were Janos Bolyai in Hungary and Nikolai Lobachevsky in Russia. Around 1830 they discovered this geometry independently and became its first “true believers.” The richness and coherence of their results convinced them that they had discovered a new geometric world, as real as the world of mainstream geometry and not needing its support. In a sense, they were right, but in their enthusiasm, they failed to notice another new geometric theory that could have been a valuable ally. Gauss’s Disquisitiones generales circa superficies curvas (General investigations on curved surfaces) was published in 1827, but neither Bolyai, Lobachevsky, or Gauss noticed that it gives models of non-Euclidean geometry, at least in small regions. The fundamental concept of Gauss’s surface theory is the curvature, a quantity that is positive (and constant) for a sphere, zero for the plane and cylinder, and negative for surfaces that are “saddle-shaped” in the neighborhood of each point. In the Disquisitiones, Gauss investigated the relationship between the curvature of a surface and the behavior of its geodesics, which are its curves of shortest length and hence its “lines.” He found, for example, that a geodesic triangle has • angle sum > π on a surface of positive curvature, • angle sum π on a surface of zero curvature, • angle sum < π on a surface of negative curvature. Moreover, if the curvature is constant and nonzero, then, in any geodesic triangle, (angle sum −π ) is proportional to area. These results must have reminded Gauss of things he already knew in non-Euclidean geometry, so it is surprising that he failed to capitalize on them.
8.9 Discussion
205
Close encounters between the actual and the hypothetical The near agreement between geometry on surfaces of constant negative curvature and non-Euclidean geometry was the first of several close encounters over the next few decades. But usually the actual and hypothetical geometries passed each other like ships in a thick fog. For example, in the late 1830s, the German mathematician Ferdinand Minding worked out the formulas of negative-curvature trigonometry. He found that they are like those of spherical trigonometry, but with hyperbolic functions in place of circular functions. At about the same time (and in the same journal!), Lobachevsky showed that the same formulas hold for triangles in his non-Euclidean plane. This would have been a nice time to introduce the name “hyperbolic geometry” for the non-Euclidean geometry of constant negative curvature, but apparently neither Minding nor Lobachevsky realized that they might have been talking about the same thing. Perhaps they were aware of a difficulty with the known surfaces of negative curvature: They are incomplete in the sense that their “lines” cannot be extended indefinitely. Hence, they fail to satisfy Euclid’s second axiom for lines. The simplest surface of constant negative curvature is called the pseudosphere (somewhat misleadingly, because constant curvature is about all it has in common with the sphere). It is more accurately known as the tractroid, because it is the surface of revolution of the curve known as the tractrix. The defining property of the tractrix is that its tangent has constant length a between the curve and the x-axis (left half of Figure 8.15).
a a
Figure 8.15: The tractrix and the tractroid
206
8 Non-Euclidean geometry
It is an unavoidable consequence of this definition that the tractrix has a singularity where the tangent becomes perpendicular to the x-axis. The tractroid likewise has an edge (like the rim of a trumpet), beyond which it cannot be smoothly continued. Hence, geodesics on the tractroid cannot be continued in both directions. In fact, the only geodesics on the tractroid that are infinite in even one direction are the rotated copies of the original tractrix. This problem is typical of what happens when one tries to construct a complete surface of constant negative curvature in ordinary space. The task was eventually shown to be impossible by Hilbert in 1901, but an obstacle to the construction of such surfaces was sensed much earlier. In 1854, Gauss’s student Bernhard Riemann showed a way round the obstacle by proposing an abstract or intrinsic definition of curved spaces— one that does not require a “flat” space to contain the “curved” one. This idea made it possible to define a complete surface, or indeed a complete n-dimensional space, of constant negative curvature. Riemann did exactly this, but once again non-Euclidean geometry sailed by unnoticed, as far as we know. (The elderly Gauss was very moved by Riemann’s account of his discoveries. Whether he saw in them a vindication of non-Euclidean geometry, we will probably never know.) Another close encounter occurred in 1859, when Arthur Cayley developed the concept of distance in projective geometry. He found that there is an invariant length for certain groups of projective transformations, such as those that map the circle into itself. In effect, he had discovered a model of the non-Euclidean plane, but he did not notice that his invariant length had the same properties as non-Euclidean length. Despite the efforts of Bolyai and Lobachevsky, non-Euclidean geometry remained an obscure subject until the 1860s.
Models of non-Euclidean geometry Riemann died in 1859, and his ideas first bore fruit in Italy, where he had spent a lot of time in his final years. His most important successor was Eugenio Beltrami, who in 1868 finally brought non-Euclidean geometry and negative curvature together. Beltrami’s first discovery, in 1865, established the special role of constant curvature in geometry: The surfaces of constant curvature are precisely those that can be mapped to the plane in such a way that geodesics go to straight lines. The simplest example is the sphere, whose geodesics are great circles, the intersections of the sphere with the planes through
8.9 Discussion
207
its center. Great circles can be mapped to straight lines by projecting the sphere onto the plane from its center (Figure 8.16).
O
Figure 8.16: Central projection of the sphere The geodesic-preserving map of the tractroid sends it to a wedge-shaped portion of the unit disk. The tractrix curves on the tractroid go to line segments ending at the sharp end of the wedge (Figure 8.17).
−→
Figure 8.17: Geodesic-preserving map of the tractroid Although this map preserves “lines,” it certainly does not preserve length. Each tractrix curve has infinite length; yet it is mapped to a finite line segment in the disk. The appropriate length function for the disk assigns a “pseudodistance” to each pair of points, equal to the geodesic distance between the corresponding points on the tractroid. We do not need the formula here; the important thing is that pseudodistance makes sense on
208
8 Non-Euclidean geometry
the whole open disk, that is, for all points inside the boundary circle. The curve of shortest pseudodistance between any two points in the open disk is the straight line segment between them, and the pseudodistance between any point and the boundary is infinite. In 1868, Beltrami realized that this abstraction and extension of the tractroid is an interpretation of the non-Euclidean plane: a surface in which there is a unique “line” between any two points, “lines” are infinite, and the non-Euclidean parallel hypothesis is satisfied. Figure 8.18 shows why: Many “lines” through the point P do not meet the “line” L .
P
L
Figure 8.18: Why the non-Euclidean parallel hypothesis holds Beltrami wrote two epic papers on models of non-Euclidean geometry in 1868, and English translations of them may be found in my book Sources of Hyperbolic Geometry. The first paper arrives at the non-Euclidean plane as an extension of the tractroid through the idea of “unwinding” infinitely thin sheets wrapped around it. (The dotted paths in the right half of Figure 8.17, all converging to the endpoint of the wedge, are the limit circles traced by circular sections of the tractroid as they unwind.) Beltrami was at pains to be as concrete as possible, because Riemann’s ideas were not well understood or accepted in 1868. However, at the end of the paper, Beltrami foreshadows the more abstract and general approach he intends to take in his second paper: where the most general principles of non-Euclidean geometry are considered independently of their possible relations with ordinary geometric entities. In the present work we have been
8.9 Discussion
209
interested mainly in offering a concrete counterpart of abstract geometry; however, we do not wish to omit a declaration that the validity of the new order of concepts does not depend on the possibility of such a counterpart. In the second paper, Beltrami vindicates this ringing endorsement of Riemann’s ideas with whole families of models of non-Euclidean geometry in any number of dimensions. Among them is the half-plane model used in this chapter, and its generalization to three dimensions, the “half-space model.” The half-space model has • “points” that are the points (x, y, z) ∈ R3 with z > 0, • “lines” that are the vertical Euclidean half lines in R3 and the vertical semicircles with centers on the plane z = 0, • “planes” that are the vertical Euclidean half planes in R3 and hemispheres with centers on z = 0. It turns out that non-Euclidean distance on a plane z = a is a constant multiple of Euclidean distance. This surprising result gives probably the simplest proof of a result first discovered by Friedrich Wachter, a member of Gauss’s circle, in 1816: Three-dimensional non-Euclidean geometry contains a model of the Euclidean plane. Another model of the hyperbolic plane, discovered by Beltrami, is the conformal disk model. It is like the half plane in being angle-preserving, but unlike it in being finite. Its “points” are the interior points of the unit disk (the points z with |z| < 1, if we work in the plane of complex numbers), and its “lines” are circular arcs perpendicular to the unit circle. Figure 8.19, which is the original M. C. Escher picture Circle Limit I, can be viewed as a picture of the conformal disk model. The fish are arranged along “lines,” and they are all of the same hyperbolic length. As mentioned in connection with Figure 8.10, the transformed Circle Limit I, the function z →
1 − zi z−i
maps the conformal disk model onto the half-plane model. It should be stressed that all models of non-Euclidean geometry, in a given dimension, are isomorphic to the half-space model. For example, models of the non-Euclidean plane satisfy Hilbert’s axioms (Section 2.9)
210
8 Non-Euclidean geometry
Figure 8.19: The conformal disk model with the parallel axiom replaced by the non-Euclidean parallel hypothesis. And Hilbert in his Grundlagen showed that the “lines” satisfying these axioms have “ends” that behave like the points of RP1 . Thus, any nonEuclidean plane is essentially the same as the half plane discussed in this chapter, so we can call it the non-Euclidean plane or the hyperbolic plane.
Non-Euclidean reality In Beltrami’s original model, the open disk in which “lines” are line segments ending on the unit circle, isometries map Euclidean lines to Euclidean lines, and so they are projective maps. For this reason, the model is often called the projective disk. It can also be constructed by methods of projective geometry, and indeed this is essentially what Cayley did in 1859. The first to connect all the dots between projective and non-Euclidean geometry was Klein in 1871. An English translation of his paper may be found in Sources of Hyperbolic Geometry. Although Klein had only to fill a few technical gaps, it was he who first made the important conceptual point that a model of non-Euclidean geometry ensures that the nonEuclidean parallel hypothesis is not contradictory. Hence, Euclid’s parallel axiom does not follow from his other axioms.
8.9 Discussion
211
In 1872, Klein also made the great advance of linking geometries to groups of transformations. This link gives a deeper reason for the presence of non-Euclidean geometry in projective geometry: The real projective line and the non-Euclidean plane have isomorphic groups of transformations. The group of the non-Euclidean plane was first described explicitly by the French mathematician Henri Poincar´e in 1882, along with its interpretation as the group of M¨obius transformations of the half plane.The relevant parts of his work may also be found in Sources of Hyperbolic Geometry. Poincar´e became interested in non-Euclidean geometry when he noticed that some functions of a complex variable have non-Euclidean periodicity. An ordinary periodic function, such as cos x, has Euclidean periodicity in the sense that its values repeat when x undergoes the Euclidean translation x → x + 2π . A complex function can have non-Euclidean periodicity, and one example is the modular function j(z). Its definition is too long to explain here, but its periodicity is simple: The values of j(z) repeat under the M¨obius transformations z → z + 1 and z → −1/z. As we know, these are isometries of the half plane. If one applies them over and over, to the lines x = 0, x = 1, and the unit semicircle, they produce the non-Euclidean regular tessellation shown in Figure 8.20.
−1
0
1
Figure 8.20: The modular tessellation The modular function and its periodicity were already part of mathematical reality, having been known to Gauss and others since early in the 19th century. But Poincar´e was the first to see its non-Euclidean symmetry. He used non-Euclidean geometry to study large classes of functions whose behavior had until then seemed intractable. Poincar´e was also the first to view the half plane as an extension of the real projective line, as we have
212
8 Non-Euclidean geometry
done in this chapter. In fact, he went much further, noticing that the halfspace model of non-Euclidean space is a natural extension of the complex projective line CP1 = C ∪ {∞}. Just as the real projective line R ∪ {∞} comes with the linear fractional transformations x →
ax + b , cx + d
where a, b, c, d ∈ R and ad − bc = 0,
the complex projective line C ∪ {∞} comes with the linear fractional transformations z →
az + b , cz + d
where a, b, c, d ∈ C and ad − bc = 0.
And just as the linear fractional transformations of R ∪ {∞} extend to M¨obius transformations of the half plane, the linear fractional transformations of C ∪ {∞} extend to M¨obius transformations of the half space, for which there is likewise an invariant non-Euclidean distance, and the non-Euclidean “lines” and “planes” mentioned above. It is a great advantage to have a concept of distance, even if the distance is non-Euclidean and one needs an extra dimension to acquire it. By passing to the third dimension, Poincar´e could understand transformations of C whose behavior is almost incomprehensible when viewed in the plane. Understanding comes by viewing these transformations as compressed versions of isometries of non-Euclidean space, which behave quite simply (like isometries of the half plane). Thus, expanding from a projective line to a non-Euclidean space is not just an interesting theoretical possibility—it is sometimes the best way to understand the mysteries of projection.
References Artmann, B.: Euclid: The Creation of Mathematics, Springer-Verlag, 1999. Behnke, H. et al.: Fundamentals of Mathematics, Volume II. Geometry, MIT Press, 1974. Brieskorn, E. and Kn¨orrer, H.: Plane Algebraic Curves, Birkh¨auser, 1986. Cayley, A.: Mathematical Papers, Cambridge University Press, 1889–1898. Coxeter, H. S. M.: Introduction to Geometry, Wiley, 1969. Descartes, R.: The Geometry, Dover, 1954. Ebbinghaus, H.-D. et al.: Numbers, Springer-Verlag, 1991. Euclid: The Thirteen Books of Euclid’s Elements, Edited by Sir Thomas Heath, Cambridge University Press, 1925. Reprinted by Dover, 1954. Euclid: Elements, Edited by Dana Densmore, Green Lion Press, 2002. Gauss, C. F.: Disquisitiones generales circa superficies curvas, Parallel Latin and English text in Dombrowski, P.: 150 Years after Gauss’ “Disquisitiones generales circa superficies curvas,” Soci´et´e Math´ematique de France, 1979. Hartshorne, R.: Geometry: Euclid and Beyond, Springer-Verlag, 2000. 213
214
References
Hilbert, D.: Foundations of Geometry, Open Court, 1971. Hilbert, D. and Cohn-Vossen, S.: Geometry and the Imagination, Chelsea, 1952. Kaplansky, I.: Linear Algebra and Geometry, a Second Course, Allyn and Bacon, 1969. McKean, H. and Moll, V.: Elliptic Curves, Cambridge University Press, 1997. Pappus of Alexandria: Book 7 of the Collection, Parts 1 and 2, Edited with translation and commentary by Alexander Jones, Springer-Verlag, 1986. Saccheri, G.: Euclides ab omni naevo vindicatus, English translation, Open Court, 1920. Snapper, E. and Troyer, R. J.: Metric Affine Geometry, Academic Press, 1971. Reprinted by Dover, 1989. Stillwell, J.: Geometry of Surfaces, Springer-Verlag, 1992. Stillwell, J.: Sources of Hyperbolic Geometry, American Mathematical Society, 1996. Veblen, O. and Young, J. W.: Projective Geometry, Ginn and Company, 1910. Wright, L.: Perspective in Perspective, Routledge and Kegan Paul, 1983.
Index 1-sphere, 158 120-cell, 173 2-sphere, 155 24-cell, 166, 173 3-sphere, 167 600-cell, 173 addition algebraic properties, 42 commutative law, 68 of area, 27 of lengths, 3, 42 of ordered pairs, 65 of ordered triples, 68 of vectors, 66 projective, 129 vector, 66, 172 parallelogram rule, 67 affine geometry, 150 theorems of, 150 map, 170 transformation, 150 view of cube, 172 Alberti, Leon Battista, 90 alternativity, 140 angle, 43 alternate interior, 22 and inner product, 68 and slope, 55 bisection, 6, 7 congruence, 44 division faulty, 9 in a circle, 36
in a semicircle, 36, 79 in half plane, 176, 186 infinitesimal view, 187 of regular n-gon, 23 right, 1 sum and area, 190 of convex n-gon, 23 of quadrilateral, 23 of triangle, 22 vertically opposite, 27 antipodal map, 156 point, 156, 168 Archimedean axiom, 45 area, 26 and angle sum, 190 as rectangle, 28 equality Greek concept, 28 Euclid’s concept, 27 Euclid’s theory of, 42 non-Euclidean, 189 of geodesic triangle, 204 of parallelogram, 29 of spherical triangle, 190 of triangle, 29 arithmetic axioms for, 43 of segments, 42 projective, 128 arithmetization, 46, 117 Artmann, Benno, 18 ASA, 20, 24, 56
215
216 and existence of parallels, 22 associative law for projective addition and scissors theorem, 136 for projective multiplication and Desargues theorem, 136 of multiplication, 117 axiom Archimedean, 45 Dedekind, 45 Desargues, 117 of circle intersection, 45 of unique line, 22 in R2 , 50 Pappus, 117 parallel, 5, 20, 21 and foundations of geometry, 174 in R2 , 50 modern, 22 Playfair’s, 22 SAS, 24 axioms congruence, 24, 44 construction, 2, 5 Euclid’s, 47 model of, 50 field, 87, 113, 117, 120 for arithmetic, 43 for non-Euclidean plane, 209 for projective geometry, 43 geometric, 5 Hilbert, 42, 47, 86, 87 list of, 43 incidence, 42, 43, 95 order, 42, 44 projective plane, 94 models of, 95 vector space, 86
Index bisection of angle, 6, 7 of line segment, 6 Blake, William, 3 Bolyai, Janos, 204 Brieskorn, Egbert, 63
C, 85 as a field, 114 as a plane, 85 Cauchy–Schwarz inequality, 80 algebraic proof, 82 Cayley, Arthur, 137 and quaternions, 173 discovery of octonions, 141 invariant length, 206 projective disk, 210 center of mass, 72 centroid, 71 as intersection of medians, 72 as vector average, 72 of tetrahedron, 74 circle angles in, 36 equation of, 52 intersection axiom, 45 non-Euclidean, 196 Cohn-Vossen, Stefan, 19 coincidence, 121 Desargues, 121 in projective arithmetic, 128, 133 in tiled floor, 122 Pappus, 121 common notions, 26 commutative law, 68 for projective addition and Pappus, 135 for projective multiplication and Pappus, 134 barycenter, 72 for vector addition, 68 Beltrami, Eugenio, 206 of multiplication, 117 models of non-Euclidean plane, 208 fails for quaternions, 137 theorem on constant curvature, 206 compass, 1 betweenness, 43 and projective addition, 129
Index axiom, 2 completeness of line, 45 complex conjugation, 178 concurrence of altitudes, 75 of medians, 73 of tetrahedron, 74 of perpendicular bisectors, 77 configuration Desargues, 13 little Desargues, 119 Pappus, 12 congruence, 43 of triangles, 24 axioms, 24, 44 of angles, 44 of line segments, 44 constructibility, 46 algebraic criterion, 55 constructible figures, 1 numbers, 45 operations, 41 construction axioms, 2, 5 bisection, 6 by straightedge alone, 88 of tiling, 92 by straightedge and compass, 1 equilateral triangle, 42 of irrational length, 11 of parallel, 8 of perpendicular, 7 of product of lengths, 10 of quotient of lengths, 10 of regular n-gon, 5 of regular 17-gon, 19 of regular pentagon, 18, 41 of regular polyhedra, 18 of right-angled triangle, 37 of square, 9 of given area, 33, 38 of square root, 38, 40
217 of square tiling, 9 coordinates, 46 and Pappus theorem, 115 homogeneous, 98 in a field, 116 in Pappian planes, 120 in projective geometry, 117 of a point in the plane, 47 cosine, 56 and inner product, 77 formula for inner product, 78, 86 function, 77 rule, 75, 78 and Pythagorean theorem, 79 costruzione legittima, 90 Coxeter, Harold Scott Macdonald, 19 CP1 and half-space model, 212 CP2 , 98 CP3 , 99 cross-ratio, 64, 108, 143 as defining invariant, 110 as fundamental invariant, 112 determination of fourth point, 110 group, 113 in half plane, 176, 191 invariance discovered by Desargues, 115 is it visible?, 109 on non-Euclidean line, 192 preserved by linear fractional functions, 108 preserved by projection, 108 transformations of, 113 cube, 18, 163 affine view, 172 of a sum, 29 curvature of space, 206 curvature of surface, 204 curve algebraic, 63 equidistant, 196 of degree 1, 63 of degree 2, 63
218
Index
D¨urer, Albrecht, 89 Einstein, Albert, 87 Dedekind axiom, 45 equation Densmore, Dana, 18 homogeneous, 98 Desargues, Girard linear, 50 configuration, 13 of circle, 52 and involutions, 199 of line, 49 and the cross-ratio, 115 of non-Euclidean line, 180 configuration equidistant spatial, 141 curves, 196 theorem, 12, 42, 115 line, 52, 76 and associative multiplication, 136, is perpendicular bisector, 76 140 point, 53 as an axiom, 117 set converse, 125 in R3 , 156 fails in Moulton plane, 121 in S2 , 156 2 fails in OP , 141 Erlanger Programm, 64 holds in HP2 , 138 Escher, Maurits, 194, 209 in space, 141 Euclid, 1 little, 119 and non-Euclidean geometry, 203 projective, 119 common notions, 26 vector version, 71 concept of equal area, 27 Descartes, Ren´e, 46 construction G´eom´etrie, 16, 41, 46 axioms, 2, 95 constructible operations, 41, 55 of equilateral triangle, 4, 6 determinant, 107, 148 of regular pentagon, 41 multiplicative property, 160 Elements, 1, 3, 4, 17 dilation, 67, 68 Heath translation, 17 and scalar multiplication, 67 geometric axioms, 5 matrix representation, 150 proofs of Pythagorean theorem, 32, 38 direction, 69 statement of parallel axiom, 21 and parallel lines, 69 theory of area, 42 relative, 69 Euclidean distance geometry, 1, 43 great-circle, 155 3-dimensional, 68 in R2 , 51 in R3 , 81 and inner product, 68 in Rn , 81 and linear algebra, 64 non-Euclidean, 191 and numbers, 43 on y-axis, 193 and parallel axiom, 20 distributive law, 138 group, 145 and Pappus theorem, 139 in vector spaces, 68 dividing by zero, 106 model of, 47 dodecahedron, 18 oriented, 145
Index via straightedge and compass, 1 inner product, 82 plane, 45, 46 space, 80
219
hyperbolic, 143, 203 Klein’s concept, 143 n-dimensional, 68 non-Euclidean, 1, 143 and parallel axiom, 174 Fano plane, 115 origin of word, 170 Fermat, Pierre de, 46 projective, 43, 87, 143 field, 113 as geometry of vision, 88 axioms, 87, 113, 117, 120 spherical, 154, 174 as coincidences, 133 is non-Euclidean, 174 in Rn , 162 vector, 65, 143 C, 114 glide reflection, 60 F2 , 114 Graves, John, 141 finite, 114 great circle, 155 of rational numbers, 114 reflection in, 156 R, 114 group with absolute value, 172 abstract, 168 foundations of geometry, 3, 12, 174, 177 continuous, 170 cyclic, 162 gaps, 45, 47, 114 isomorphism, 169 Gauss, Carl Friedrich, 204 of isometries, 64 and modular function, 211 of transformations, 64, 113, 143, construction of 17-gon, 19 144, 168 Disquisitiones generales, 204 theory, 64 generating transformations, 103 are products of reflections, 183 H, 137 of half plane, 179 half plane, 175 of RP1 , 178 generating transformations of, 179 preserve cross-ratio, 108 isomorphic to RP1 , 194 geodesic, 204 tilings, 188 geometry half space, 209 affine, 150 and CP1 , 212 algebraic, 63 Hamilton, William Rowan, 137 arithmetization of, 46 and rotations of R3 , 173 Cartesian, 46 search for number systems, 172 coordinate, 46 Harriot, Thomas, 190 differential, 174 Hartshorne, Robin, 18, 43, 56 Euclidean, 1, 20, 43, 143 on the cross-ratio, 108 3-dimensional, 68 Harunobu, Suzuki, 170 and inner product, 68 Hessenberg, Gerhard, 140 and linear algebra, 64 hexagon, 5 in vector spaces, 68 regular, 5 plane of, 45 construction, 6 foundations of, 3, 12, 174, 177 tiling, 6
220 perspective view, 94 Hilbert, David, 19, 140 axioms, 42, 86, 87 for non-Euclidean plane, 209 list of, 43 constructed non-Pappian plane, 140 Grundlagen, 42, 140, 210 segment arithmetic, 42 theorem on negative curvature, 206 homogeneous coordinates, 98 horizon, 89, 100 horocycle see limit circle 198 HP2 , 138 satisfies Desargues, 138 violates Pappus, 138 hyperbolic geometry, 143, 203 plane, 210 uniqueness, 210 icosahedron, 18 identity function, 144 incidence, 42, 95 axioms, 43, 95 in projective space, 99 theorem, 118 infinity, 2 line at, 92, 100 point at, 92 projection from, 101 inner product, 65 algebraic properties, 75 and cosine, 77 and Euclidean geometry, 68 and length, 65, 75 and perpendicularity, 75 cosine formula, 78, 86 in R2 , 74 Rn , 81 positive definite, 82 intersection, 53 of circles, 4, 5, 54 axiom, 45 of lines, 54
Index invariant, 64, 110 fundamental, 112 length in half plane, 176 of group of transformations, 143 of isometry group, 143, 145 of projective transformations, 143 inversion in a circle, 179 involution, 199 irrational length, 11, √ 28 number 2, 16, 17, 47 isometry, 57 and motion, 57 composite, 144 group, 64 of half plane, 196 of non-Euclidean space, 212 of R2 , 61, 144 of R3 , 155 of S2 , 155 reason for name, 58 isomorphism, 169 between geometries, 194 between models, 209 isosceles triangle theorem, 24 Joyce, David, 18 Kaplansky, Irving, 87 Klein, Felix, 64, 143 and projective disk, 210 and the parallel axiom, 210 and transformation groups, 143, 211 concept of geometry, 143 Kn¨orrer, Horst, 63 law of cosines, 75 length addition, 3, 42 and inner product, 65, 68, 75 division, 10 in R2 , 51 irrational, 11, 28 multiplication, 10, 42
Index non-Euclidean, 176 product and Thales’ theorem, 10 rational, 11 subtraction, 3 limit circle, 198 limit rotation, 198 line, 2 as fixed point set, 183 at infinity, 92, 100 broken, 120 completeness of, 45 defined by linear equation, 50 equation of, 49 equidistant, 52, 76 is perpendicular bisector, 76 non-Euclidean, 176 is infinite, 195 number, 47 of Moulton plane, 120 perpendicular, 7 projective, 97 algebraic definition, 107 modelled by circle, 97, 101 projection of, 101 real, 97 segment, 2, 43 bisection, 6 congruence of, 44 n-section, 8 slope of, 48 linear algebra, 64, 65 equation, 50 independence, 69 transformation, 143, 146 inverse, 148 matrix representation, 148 preserves parallels, 147 preserves straightness, 147 preserves vector operations, 146 linear fractional functions, 104 are realized by projection, 104
221 behave like matrices, 107 characterization, 111 defining invariant of, 110 generators of, 108 on real projective line, 107 orientation-preserving, 182 preserve cross-ratio, 108 linear fractional transformations see linear fractional functions 104 lines asymptotic, 203 parallel, 1, 21 and direction, 69 little Desargues theorem, 119 and alternative multiplication, 140 and projective addition, 129 and tiled-floor coincidence, 123 fails in Moulton plane, 121 implies little Pappus, 135 little Pappus theorem, 135 Lobachevsky, Nikolai Ivanovich, 204 hyperbolic formulas, 205 logic, 5 magnification see dilation 13 matrix, 83 determinant of, 148 of linear fractional function, 107 of linear transformation, 147 product, 148 rotation, 83 McKean, Henry, 63 medians, 72 concurrence of, 73 midpoint, 71 Minding, Ferdinand, 205 Minkowski space, 87 M¨obius transformations, 182 and cross-ratio, 192 generating, 179 preserve angle, 186 preserve non-Euclidean lines, 184 restriction to RP1 , 182 model
222
Index
of Euclid’s axioms, 50 invariance, 184 of Euclidean plane geometry, 47 uniqueness, 177 of non-Euclidean geometry, 204, 206 violate parallel axiom, 176 of non-Euclidean plane, 209 parallel hypothesis, 203 of non-Euclidean space, 209 periodicity, 211 of projective line, 97 plane, 64 of projective plane, 95 axioms, 209 of the line, 47 from projective line, 174 modular function, 211 uniqueness, 210 Moll, Victor, 63 rotation, 196 motion, 24, 46, 56 space, 209, 212 is an isometry, 57 isometries of, 212 Moufang, Ruth, 140 symmetry, 211 and HP2 , 141 translation, 196 and little Desargues theorem, 140 triangle, 188 and OP2 , 141 non-Pappian plane, 138, 141 Moulton plane, 120, 141 of Hilbert, 140 lines of, 120 numbers, 1 violates converse Desargues, 128 as coordinates, 46 violates little Desargues, 121 complex, 84 violates tiled-floor coincidence, 124 and non-Euclidean plane, 174 Moulton, Forest Ray, 128 and rotation, 84 multiplication constructible, 45 algebraic properties, 42 irrational, 47 associative law, 117 n-dimensional, 172 commutative law, 117 nonconstructible, 45 noncommutative, 137, 162 found by Wantzel, 55 of lengths, 42 prime, 5, 19 projective, 130 rational, 47 real, 45, 47 Newton, Isaac, 3 as vectors, 66 non-Euclidean area, 189 O, 141 circle, 196 octahedron, 18, 173 distance, 191 octonion projective plane, 141 is additive, 193 discovered by Moufang, 141 on y-axis, 193 satisfies little Desargues, 141 geometry, 1, 143 violates Desargues, 141 and parallel axiom, 174 octonion projective space models of, 204, 206 does not exist, 142 length, 176 octonions, 141 lines, 46, 176 OP2 , 141 satisfies little Desargues, 141 are infinite, 195 violates Desargues, 141 equations of, 180
Index order, 42 axioms, 44 ordered n-tuple, 68 ordered pair, 48 addition, 65 as vector, 65 scalar multiple, 65 ordered triple, 68 origin, 47
223
opposite sides are equal, 25 rule, 67 pentagon regular, 18 construction of, 41 perpendicular, 7 bisector, 76 construction of, 7 perspective, 88 drawing, 89 Pappus, 24 view of equally-spaced points, 91 configuration, 12 view of RP2 , 100 labeled by vectors, 71 view of tiling, 88 theorem, 12, 42, 65 by straightedge alone, 92 and commutative multiplication, π , 45 134, 140 plane and coordinates, 115 hyperbolic, 210 and distributive law, 139 Moulton, 120 and projective addition, 132 non-Euclidean, 64, 174 as an axiom, 117 non-Pappian, 138, 141 2 fails in HP , 138 number, 47 implies Desargues, 140 of Euclidean geometry, 45 projective version, 118 Pappian, 119 vector version, 70 projective, 94 parallel real number, 45 construction, 8 real projective, 95 lines, 1, 21 Playfair, John, 22 and ASA, 22 Poincar´e, Henri, 211 have same slope, 50 point at infinity, 92 in projective plane, 95 polygon parallel axiom, 20, 21, 42 regular, 5 and foundations of geometry, 174 squaring, 38 and non-Euclidean geometry, 174 polyhedra does not follow regular, 18 from other axioms, 210 polytope, 166 equivalents, 203 positive definite, 82 Euclid’s statement, 21 fails for non-Euclidean lines, 176 postulates see axioms 2 product independent of the others, 177 as group operation, 168 Playfair’s statement, 22 of functions, 168 parallelogram, 25 of lengths area of, 29 as rectangle, 16, 28 diagonals bisect, 26 vector proof, 72 by straightedge and compass, 10
224 of matrices, 148 of rotations, 157, 159 projection, 100 axonometric, 171 from finite point, 102 from infinity, 101 in Japanese art, 170 is linear fractional, 104 of projective line, 101 preserves cross-ratio, 108 projective addition, 129 arithmetic, 128 Desargues configuration, 119 disk, 210 distortion, 102 geometry, 43, 87 and non-Euclidean plane, 174 as geometry of vision, 88 axioms, 43 coordinates in, 117 reason for name, 100 line algebraic definition, 107 real, 97 little Desargues configuration, 119 multiplication, 130 Pappus configuration, 118 plane, 94 axioms, 94 complex, 98 extends Euclidean plane, 96 Fano, 115 FP2 , 113 model of axioms, 95 octonion, 141 quaternion, 138 real, 95 plane axioms models of, 99 planes, 117 reflection, 182 space, 99
Index 3-dimensional, 99 complex, 99 incidence properties, 99 real, 99, 141 three-dimensional, 167 transformations, 151 as linear transformations, 151 of RP1 , 106, 184 pseudosphere see tractroid 205 Pythagorean theorem, 20 and distance in R2 , 51 and distance in R3 , 81 and cosine rule, 79 Euclid’s first proof, 32 Euclid’s second proof, 38 in R2 , 51 Pythagoreans, 11 quaternion projective plane, 138 satisfies Desargues, 138 quaternions, 137 and the fourth dimension, 172 as complex matrices, 137, 159 noncommutative multiplication, 137, 162 opposite, 162 represent rotations, 159 quotient of lengths, 10 R, 47 as a field, 114 as a line, 47 R2 , 45 as a field, 172 as a plane, 48 as a vector space, 66 distance in, 51 isometry of, 61, 144 R3 , 95 regular polyhedra in, 173 rotation of, 156 R4 , 172 regular polytopes in, 173 rational
Index length, 11 numbers, 47 ray, 43 rectangle, 21 as product of lengths, 16 reflection and M¨obius transformations, 182 as linear transformation, 149 by generating transformations, 183 fixed point set is a line, 183 in a circle, 182 in a sphere, 182 in great circle, 156 in unit circle, 179 of half plane, 178 of R2 in x-axis, 60 in any line, 60 ordinary, 183 projective, 182 regular 17-gon, 19 hexagon, 5 n-gon, 162 pentagon, 18 polygon, 5 polyhedra, 18, 163 polytopes, 166 discovered by Schlafli, 173 relativity, 87 rhombus, 26 has perpendicular diagonals, 26 vector proof, 76 Riemann, Bernhard, 206 right angle, 1 Rn , 68 as a Euclidean space, 81 as a vector space, 68 rotation and complex numbers, 84 and multiplication by −1, 68 as linear transformation, 149 as product of reflections, 60
225 group of sphere, 157 limit, 198 matrix, 83 non-Euclidean, 196 of R2 , 57, 59 of R3 , 156, 159 of S1 , 158 of S2 , 157, 159 of tetrahedron, 163 RP1 , 107, 151 as boundary of half plane, 175 isomorphic to half plane, 194 RP2 , 95, 151 RP3 , 99, 167 and geometry of the sphere, 99 as a group, 168 S1 , 158 as a group, 169 rotation of, 158 S2 , 155 isometry of, 155 noncommuting rotations, 159 rotation of, 157 S3 , 167 as a group, 169 Saccheri, Girolamo, 203 SAS, 20, 56 statement, 24 scalar multiple, 66 scalar product see inner product 74 Schl¨afli, Ludwig, 173 scissors theorem, 126 and Desargues theorem, 126 and projective multiplication, 130, 132 little and little Desargues, 128 fails in Moulton plane, 128 semicircle angle in, 36, 79 as non-Euclidean line, 176 sine, 56 slope, 48
226 and angle, 55 and perpendicularity, 56 infinite, 49 reciprocal, 153 relative, 56 Snapper, Ernst, 87 space 3-dimensional, 68 curved, 206 Desargues theorem in, 141 Euclidean, 81 half, 209 M¨obius transformations of, 212 n-dimensional, 68 non-Euclidean, 209, 212 isometries of, 212 projective, 99 complex, 99 n-dimensional, 141 real, 99 rotations of, 159 vector, 86 real, 67 sphere 0-dimensional, 184 1-dimensional, 158, 167 2-dimensional, 155, 167 3-dimensional, 167 in R3 , 154 spherical geometry, 154 lines of, 155 triangle, 188, 190 square, 18 construction, 9 diagonal of, 15 of a sum, 27 tiling construction, 9 square root, 38 construction, 40 squaring a polygon, 38
Index the circle, 38, 45 SSS, 24 straightedge, 1 axioms, 2 surface curvature of, 204 incomplete, 205 of constant curvature, 204 tetrahedron, 18, 74, 163 centroid of, 74 rotations of, 163 as quaternions, 165 Thales, 8 theorem on right angles, 36 Thales theorem, 8, 20, 65 and product of lengths, 10 and quotient of lengths, 11 converse, 11 proof, 34 vector version, 70 theorem Desargues, 12, 42, 115 converse, 125 projective, 119 vector version, 71 intermediate value, 202 isosceles triangle, 24 little Desargues, 119 fails in Moulton plane, 121 on concurrence of altitudes, 75 on concurrence of medians, 73 Pappus, 12, 42, 65 and coordinates, 115 little, 135 projective version, 118 vector version, 70 Pythagorean, 20 and cosine rule, 79 Euclid’s first proof, 32 Euclid’s second proof, 38 scissors, 126 Thales, 8, 20, 65 proof, 34
Index vector version, 70 three reflections, 61 for half plane, 184, 199 for R2 , 62, 184 for RP1 , 184 for S2 , 156 two involutions, 199 theory of proportion, 31, 43 three-point maps existence, 111 uniqueness, 111 tiling by equilateral triangles, 6 perspective view, 94 by regular hexagons, 6 perspective view, 94 by regular n-gons, 23 of half plane, 188 perspective view, 88 by straightedge alone, 92 tractrix, 205 tractroid, 205 geodesic-preserving map, 207 geodesics on, 206 transformation, 144 affine, 150 group, 113, 144 for Euclidean geometry, 145 invariant of, 143 inverse, 144 linear, 143, 146 linear fractional, 104 M¨obius, 182 of cross-ratio, 113 of R2 , 57 projective, 64, 151 invariant of, 143 similarity, 150 stretch, 149 translation, 196 as product of reflections, 60 axis, 196 of R2 , 58
227 triangle area formula, 30 area of, 29 equilateral, 4, 18 construction, 4 tiling, 6 tiling in perspective, 94 inequality, 53 from Cauchy–Schwarz, 80 isosceles, 20, 24 non-Euclidean, 188 spherical, 188 triangles congruent, 24 similar, 13 have proportional sides, 13 trisection, 9 impossibility of, 55 Troyer, Robert, 87 unit of length, 10 vanishing point see point at infinity 92 Veblen, Oswald, 199 vector, 65 addition, 66, 172 additive inverse of, 67 algebraic properties, 66 average, 72 column, 148 scalar multiple of, 66 zero, 67 vector space, 67 axioms, 86 real, 67 Rn , 68 vertically opposite angles, 27 von Staudt, Christian, 140 Wachter, Friedrich, 209 Wantzel, Pierre, 19 and constructibility, 55 Wiener, Hermann, 140 Young, John Wesley, 199
Undergraduate Texts in Mathematics (continued from page ii)
Franklin: Methods of Mathematical Economics. Frazier: An Introduction to Wavelets Through Linear Algebra Gamelin: Complex Analysis. Gordon: Discrete Probability. Hairer/Wanner: Analysis by Its History. Readings in Mathematics. Halmos: Finite-Dimensional Vector Spaces. Second edition. Halmos: Naive Set Theory. Ha¨mmerlin/Hoffmann: Numerical Mathematics. Readings in Mathematics. Harris/Hirst/Mossinghoff: Combinatorics and Graph Theory. Hartshorne: Geometry: Euclid and Beyond. Hijab: Introduction to Calculus and Classical Analysis. Hilton/Holton/Pedersen: Mathematical Reflections: In a Room with Many Mirrors. Hilton/Holton/Pedersen: Mathematical Vistas: From a Room with Many Windows. Iooss/Joseph: Elementary Stability and Bifurcation Theory. Second edition. Irving: Integers, Polynomials, and Rings: A Course in Algebra Isaac: The Pleasures of Probability. Readings in Mathematics. James: Topological and Uniform Spaces. Ja¨nich: Linear Algebra. Ja¨nich: Topology. Ja¨nich: Vector Analysis. Kemeny/Snell: Finite Markov Chains. Kinsey: Topology of Surfaces. Klambauer: Aspects of Calculus. Lang: A First Course in Calculus. Fifth edition. Lang: Calculus of Several Variables. Third edition. Lang: Introduction to Linear Algebra. Second edition.
Lang: Linear Algebra. Third edition. Lang: Short Calculus: The Original Edition of “A First Course in Calculus.” Lang: Undergraduate Algebra. Third edition Lang: Undergraduate Analysis. Laubenbacher/Pengelley: Mathematical Expeditions. Lax/Burstein/Lax: Calculus with Applications and Computing. Volume 1. LeCuyer: College Mathematics with APL. Lidl/Pilz: Applied Abstract Algebra. Second edition. Logan: Applied Partial Differential Equations, Second edition. Logan: A First Course in Differential Equations. Lova´sz/Pelika´n/Vesztergombi: Discrete Mathematics. Macki-Strauss: Introduction to Optimal Control Theory. Malitz: Introduction to Mathematical Logic. Marsden/Weinstein: Calculus I, II, III. Second edition. Martin: Counting: The Art of Enumerative Combinatorics. Martin: The Foundations of Geometry and the Non-Euclidean Plane. Martin: Geometric Constructions. Martin: Transformation Geometry: An Introduction to Symmetry. Millman/Parker: Geometry: A Metric Approach with Models. Second edition. Moschovakis: Notes on Set Theory. Owen: A First Course in the Mathematical Foundations of Thermodynamics. Palka: An Introduction to Complex Function Theory. Pedrick: A First Course in Analysis. Peressini/Sullivan/Uhl: The Mathematics of Nonlinear Programming.
Undergraduate Texts in Mathematics Prenowitz/Jantosciak: Join Geometries. Priestley: Calculus: A Liberal Art. Second edition. Protter/Morrey: A First Course in Real Analysis. Second edition. Protter/Morrey: Intermediate Calculus. Second edition. Pugh: Real Mathematical Analysis. Roman: An Introduction to Coding and Information Theory. Roman: Introduction to the Mathematics of Finance: From Risk Management to Options Pricing. Ross: Differential Equations: An Introduction with Mathematica®. Second edition. Ross: Elementary Analysis: The Theory of Calculus. Samuel: Projective Geometry. Readings in Mathematics. Saxe: Beginning Functional Analysis Scharlau/Opolka: From Fermat to Minkowski. Schiff: The Laplace Transform: Theory and Applications. Sethuraman: Rings, Fields, and Vector Spaces: An Approach to Geometric Constructability. Sigler: Algebra. Silverman/Tate: Rational Points on Elliptic Curves.
Simmonds: A Brief on Tensor Analysis. Second edition. Singer: Geometry: Plane and Fancy. Singer: Linearity, Symmetry, and Prediction in the Hydrogen Atom Singer/Thorpe: Lecture Notes on Elementary Topology and Geometry. Smith: Linear Algebra. Third edition. Smith: Primer of Modern Analysis. Second edition. Stanton/White: Constructive Combinatorics. Stillwell: Elements of Algebra: Geometry, Numbers, Equations. Stillwell: Elements of Number Theory. Stillwell: The Four Pillars of Geometry. Stillwell: Mathematics and Its History. Second edition. Stillwell: Numbers and Geometry. Readings in Mathematics. Strayer: Linear Programming and Its Applications. Toth: Glimpses of Algebra and Geometry. Second Edition. Readings in Mathematics. Troutman: Variational Calculus and Optimal Control. Second edition. Valenza: Linear Algebra: An Introduction to Abstract Mathematics. Whyburn/Duda: Dynamic Topology. Wilson: Much Ado About Calculus.